@leadbay/mcp 0.18.1 → 0.19.0

package/dist/bin.js

@@ -5847,29 +5847,23 @@ Do not enumerate the affected leads \u2014 that's the job of \`leadbay_pull_lead
 
 ---
 
+---
+
 ## NEXT STEPS \u2014 after kicking off bulk qualification
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -6001,7 +5995,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -6093,6 +6087,8 @@ WHEN TO USE: after \`leadbay_list_campaigns\` (or when the user named a specific
 WHEN NOT TO USE: for cross-campaign pulse (use \`leadbay_list_campaigns\`); to drill into a lead's full timeline (use \`leadbay_get_lead_activities\` or \`leadbay_research_lead_by_id\`); to log outreach (\`leadbay_report_outreach\`).
 
 **Response**: \`{items, pagination, summary, _meta}\`. Use the \`summary\` for the one-line headline; use \`items\` for the per-lead table.
+
+---
 `;
     leadbay_clear_selection = `Clear the user's transient selection.
 
@@ -6300,7 +6296,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -6337,7 +6333,7 @@ Examples that should NOT invoke this tool (sound similar, route elsewhere):
 
 \`queued\` \u2192 \u2705 "Queued <N> extra leads on lens <id>. Pull in ~30s." Do NOT
 list \`accepted_seeds\`; they're internal.
-\`quota_exceeded\` \u2192 render three options via \`ask_user_input_v0\` (smaller
+\`quota_exceeded\` \u2192 render three options via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`) (smaller
 count / wait until reset / upgrade).
 \`refresh_in_progress\` \u2192 "lens is filling, retry in a minute".
 \`no_valid_seeds\` \u2192 silently re-call \`leadbay_seed_candidates\`, retry once.
@@ -6355,7 +6351,7 @@ Queue an additive extra-refill on a lens \u2014 more leads on the same criteria,
 **Status envelope (translated from raw API errors so the agent routes on \`status\`).**
 
 - \`status: "queued"\` \u2014 fill is queued. \`accepted_seeds\` lists IDs that passed validation. NEXT STEP: call \`leadbay_pull_leads\` in ~30s.
-- \`status: "quota_exceeded"\` \u2014 daily LENS_EXTRA_REFILL hit. Response carries \`quota: {used_today, resets_at}\` + a \`message\` to surface. **Render three options via \`ask_user_input_v0\`**: (1) smaller \`extra_count\`, (2) wait until \`resets_at\`, (3) upgrade plan (TIER1=150, TIER2=1000). Do NOT silently retry.
+- \`status: "quota_exceeded"\` \u2014 daily LENS_EXTRA_REFILL hit. Response carries \`quota: {used_today, resets_at}\` + a \`message\` to surface. **Render three options via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`)**: (1) smaller \`extra_count\`, (2) wait until \`resets_at\`, (3) upgrade plan (TIER1=150, TIER2=1000). Do NOT silently retry.
 - \`status: "refresh_in_progress"\` \u2014 a refresh or extra-refill is already running. Tell the user to wait and call \`leadbay_pull_leads\` in ~30s.
 - \`status: "no_valid_seeds"\` \u2014 seeds went stale. Silently re-call \`leadbay_seed_candidates\` and retry once; only surface to the user if the second attempt also fails.
 
@@ -6370,27 +6366,19 @@ This tool MUTATES state. The caller (agent or human-in-the-loop) is responsible
 
 ## NEXT STEPS \u2014 after \`leadbay_extend_lens\`
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -6533,7 +6521,7 @@ Never link a person's name to the company's LinkedIn page (and vice versa) \u201
 
 Open with **one short intro sentence** in chat ("Five lead visits across NYC for your trip next week \u2014 three in Midtown, plus Long Island and one in NJ.") and then invoke the widget, then the chat-side list above. **No markdown table.**
 
-**After the widget renders, end the turn with the NEXT STEPS surface** \u2014 not with a prose question. See "GATE \u2014 PREFER BUILT-IN HOST WIDGETS" below: surface 2\u20134 mutually-exclusive moves via \`ask_user_input_v0\` if the host exposes it, else as a short bulleted list. "Want me to plot these on a map or jump to outreach for Atlas?" is exactly the prose pattern to AVOID \u2014 it's a \`single_select\` with two options.
+**After the widget renders, end the turn with the NEXT STEPS surface** \u2014 not with a prose question. See "GATE \u2014 PREFER BUILT-IN HOST WIDGETS" below: surface 2\u20134 mutually-exclusive moves via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`) if the host exposes it, else as a short bulleted list. "Want me to plot these on a map or jump to outreach for Atlas?" is exactly the prose pattern to AVOID \u2014 it's a \`single_select\` with two options.
 
 ## RENDER \u2014 fallback for hosts without \`places_map_display_v0\`
 
@@ -6564,7 +6552,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -6709,27 +6697,19 @@ Defer the full list of imported leads to \`leadbay_pull_leads\` or \`leadbay_res
 
 ## NEXT STEPS \u2014 after an import
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -6782,27 +6762,19 @@ Defer the full list of imported leads to \`leadbay_pull_leads\` or \`leadbay_res
 
 ## NEXT STEPS \u2014 after an import
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -6942,6 +6914,8 @@ WHEN TO USE: the user wants the cross-campaign pulse \u2014 what cohorts are in
 WHEN NOT TO USE: for per-lead progression inside ONE campaign (use \`leadbay_campaign_progression\`); to create a campaign (\`leadbay_create_campaign\`); to add leads to one (\`leadbay_add_leads_to_campaign\`).
 
 **Response**: \`{campaigns: CampaignWithStats[], _meta}\`. Sort by \`updated_at desc\` when rendering \u2014 recency is the manager's natural lens.
+
+---
 `;
     leadbay_list_lenses = `List all available Leadbay lenses (saved lead-search configurations). Each lens defines a different target market or buyer segment. The lens with \`is_last_active=true\` is used by default for lead discovery.
 
@@ -7068,7 +7042,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -7107,27 +7081,19 @@ render an empty table.
 
 ## NEXT STEPS \u2014 after \`leadbay_my_lenses\`
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -7205,7 +7171,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -7220,27 +7186,19 @@ ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We
 
 ## NEXT STEPS \u2014 after \`leadbay_new_lens\`
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -7309,7 +7267,7 @@ email. Do NOT paste the email body into chat prose alongside.
 
 ---
 
-Prepare a single-lead outreach brief: the full \`lead\` block (score, \`split_ai_summary\`, \`location\`, \`size\`, \`phone_numbers\`, \`website\`, \`description\`, \`social_urls\`, \`social_presence\`), the \`recommended_contact\` (always in the post-enrichment shape \u2014 \`contact_id\`, \`first_name\`, \`last_name\`, \`job_title\`, \`email\`, \`phone_number\`, \`linkedin_page\`, \`is_org_contact\` \u2014 with nulls where data isn't yet enriched), \`additional_contacts_count\` (other contacts at this company), and an \`enrichment\` block describing async state.
+Prepare a single-lead outreach brief: the full \`lead\` block (score, \`split_ai_summary\`, \`location\`, \`size\`, \`phone_numbers\`, \`website\`, \`description\`, \`social_urls\`, \`social_presence\`), the \`recommended_contact\` (always in the post-enrichment shape \u2014 \`contact_id\`, \`first_name\`, \`last_name\`, \`job_title\`, \`email\`, \`phone_number\`, \`linkedin_page\`, \`is_org_contact\` \u2014 with nulls where data isn't yet enriched), \`additional_contacts_count\`, and an \`enrichment\` block describing async state.
 
 Optionally trigger contact enrichment in-flight with \`enrich:true\`. Enrichment is async (~60s). **Self-polling pattern (no separate tool needed):** re-call \`leadbay_prepare_outreach(leadId)\` without \`enrich\`; check \`enrichment.complete\`. When \`complete: true\`, the recommended contact now carries \`email\` and/or \`phone_number\`.
 
@@ -7326,46 +7284,9 @@ WHEN NOT TO USE: across many leads \u2014 use leadbay_enrich_titles for bulk; fo
 
 ## RENDER \u2014 host-native message composer is the PRIMARY surface
 
-\`message_compose_v1\` is the canonical surface for outreach drafts on this tool. The composer gives the user inline edit + send affordances and beats any prose code-fenced draft. **If the host exposes \`message_compose_v1\`, route every draft through it.** Don't paste email body or call-opener body into chat prose alongside \u2014 the composer IS the visual.
-
-Above the composer, emit ONE short markdown context paragraph: the lead's score callout + sector fit + recommended-contact name (LinkedIn-markdown-linked) + bare phone/email pills. That gives the user the "why this lead" context. The composer below carries the actionable draft.
-
-**Single draft:**
+Route every draft through \`message_compose_v1\` (Claude's email composer). Above it, emit ONE short markdown context paragraph: score callout + sector fit + linked contact name + bare phone/email pills. Do NOT paste the email body into chat prose alongside \u2014 the composer IS the visual.
 
-\`\`\`
-message_compose_v1({
-  kind: "email",
-  summary_title: "Outreach to <Contact Name> at <Company>",
-  variants: [{
-    label: "Lead with the M&A signal",
-    subject: "<one-line subject \u2014 references the angle>",
-    body: "<5-8 sentence email; salesperson voice; references signal + a clear next step>"
-  }]
-})
-\`\`\`
-
-**Strategic options (preferred when split_ai_summary surfaces multiple angles):**
-
-\`\`\`
-message_compose_v1({
-  kind: "email",
-  summary_title: "Three angles for <Company> outreach",
-  variants: [
-    { label: "Push for alignment",  subject: "...", body: "..." },
-    { label: "Reference the M&A signal", subject: "...", body: "..." },
-    { label: "Soft intro \u2014 peer reference", subject: "...", body: "..." }
-  ]
-})
-\`\`\`
-
-Constraints:
-- **Labels describe STRATEGY, not tone.** "Push for alignment", "Reference M&A signal", "Lead with peer reference" \u2014 not "Friendly" / "Formal" / "Aggressive".
-- **2\u20133 variants when strategic options are clearly distinct.** One variant when you have a single best-angle draft.
-- Subject required for \`kind: "email"\`. Phone/call openers use \`kind: "other"\` with the opener in \`body\`.
-
-The composer becomes the single visual. **Don't also paste the email body into chat prose** \u2014 that's just noise next to the composer.
-
-For phone-only contacts (no email enriched), use \`kind: "other"\` with a 60-second call opener.
+Variant shape: 1\u20133 entries. Labels describe **strategy** ("Push for alignment", "Reference the M&A signal", "Soft intro \u2014 peer reference"), not tone. \`kind: "email"\` requires \`subject\`; phone/call openers use \`kind: "other"\` with the opener in \`body\`.
 
 ## GATE \u2014 PREFER BUILT-IN HOST WIDGETS
 
@@ -7377,7 +7298,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -7458,27 +7379,19 @@ When the response carries \`social_urls\` (the post-fix multi-platform URL block
 
 ## NEXT STEPS \u2014 after the outreach brief
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -7498,7 +7411,6 @@ Offer 2\u20133 follow-ups. Choose based on enrichment state + available channels
 | User reports they reached out                   | "Log this outreach \u2014 creates prospecting action + outcome"    | leadbay_report_outreach(leadId, contact_id, ...)       |
 | User adds context for next time                 | "Save a note on the contact or company"                       | leadbay_add_note                                       |
 | After a successful exchange                     | "Update qualification answers based on what you learned"      | leadbay_answer_clarification                           |
-
 The "log outreach" step is the most-important follow-up \u2014 it closes the loop and populates history for the next \`leadbay_prepare_outreach\` call. Detect intent from natural language: "I sent the email", "she didn't pick up", "left a voicemail", "they responded yes/no", etc.
 `;
     leadbay_preview_bulk_enrichment = `Preview a bulk-enrichment cost given a set of job titles applied to the current selection. Returns \`{selected_leads, enriched_contacts, enrichable_contacts, title_suggestions, auto_included_titles, previously_enriched_titles}\`. \`previously_enriched_titles\` is a newer field (in prod soon) \u2014 when present, the agent can recommend repeating those titles for new leads.
@@ -7662,27 +7574,19 @@ When the response carries \`social_urls\` (the post-fix multi-platform URL block
 
 ## NEXT STEPS \u2014 after the follow-ups table
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -7703,7 +7607,6 @@ Always include at least one filter-modification offer (users think in filters: b
 | User wants to defer a lead                    | "Snooze [Company] for 3 / 6 / 12 months"                 | leadbay_set_pushback({ lead_ids:[leadId], status:"3" })                                            |
 | User completed outreach mid-flow              | "Log the outreach + record the outcome"                  | leadbay_report_outreach                                                                            |
 | Discovery mode might fit better               | "Looking for NEW leads instead? Switch to discovery."    | leadbay_pull_leads                                                                                 |
-
 Always offer at least one of: prep outreach, refilter, pushback. Pushback is the canonical way to honor "not now" / "next quarter" \u2014 leads with active pushback are excluded from this view until expiry.
 `;
     leadbay_pull_leads = `## WHEN TO USE
@@ -7830,27 +7733,19 @@ When the response carries \`social_urls\` (the post-fix multi-platform URL block
 
 ## NEXT STEPS \u2014 after rendering the pull_leads table
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -7860,6 +7755,7 @@ Pick 2\u20133 items below based on what was actually observed in the response. T
 
 | Observation                                                | Suggest                                                      | Calls                                                  |
 |------------------------------------------------------------|--------------------------------------------------------------|--------------------------------------------------------|
+| \u2265 5 leads returned (any batch)                             | "Build an interactive lead triage board for this batch"      | emit antArtifact from data in hand (do NOT re-call leadbay_pull_leads) |
 | \`has_more == true\`                                         | "Pull the next page (page N+1 of M)"                         | leadbay_pull_leads(page = current + 1, lensId = pinned)|
 | \u2265 3 rows have \`qualification_summary.answered == 0\`        | "Deepen AI qualification on the rows without \u2756 caps"         | leadbay_bulk_qualify_leads(leadIds=[\u2026])                |
 | User points at a single row                                | "Research [Company] in depth"                                | leadbay_research_lead_by_id(leadId)                    |
@@ -7870,7 +7766,7 @@ Pick 2\u20133 items below based on what was actually observed in the response. T
 | Top row has contacts but no phone/email                    | "Order contact enrichment to surface email/phone first"      | leadbay_enrich_titles(...) or leadbay_prepare_outreach(leadId, enrich:true) |
 | \`computing_scores == true\` or \`computing_wishlist == true\` | "Scores are still being computed \u2014 re-pull in ~30s"          | leadbay_pull_leads (retry with same lensId)            |
 | User wants a narrower / wider audience                     | "Adjust the lens filters (sector / size)"                    | leadbay_adjust_audience(...)                           |
-
+| Phase 4 research was run (\`research_lead_by_id\` called) AND top contacts lack direct email/phone | "Enrich contacts on [Lead1], [Lead2] to get direct emails and phone numbers" | leadbay_enrich_contacts(leadId, contactId) \u2014 ONE call per contact (the tool takes a single leadId + contactId, never a list) |
 If nothing in the menu applies cleanly, suggest only "pull next page" and "research a specific lead in depth" \u2014 never invent a tool that doesn't exist.
 `;
     leadbay_qualify_lead = `Trigger AI qualification for a single lead (web fetch + AI rescore). The operation is asynchronous \u2014 results take ~60s. \`forceFetch:true\` re-runs even if recent data exists.
@@ -8184,27 +8080,19 @@ When the response carries \`social_urls\` (the post-fix multi-platform URL block
 
 ## NEXT STEPS \u2014 after the research card
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -8249,6 +8137,7 @@ out?"\`
 
 | Observation                                            | Suggest                                                  | Calls                                                          |
 |--------------------------------------------------------|----------------------------------------------------------|----------------------------------------------------------------|
+| Lead is clearly not a fit (wrong industry, too small)  | "Dislike this lead"                                      | leadbay_dislike_lead({ leadId })                               |
 | User is done with this lead                            | "Back to the inbox"                                      | leadbay_pull_leads                                             |
 `;
     leadbay_research_lead_by_name_fuzzy = `## WHEN TO USE
@@ -8372,27 +8261,19 @@ When the response carries \`social_urls\` (the post-fix multi-platform URL block
 
 ## NEXT STEPS \u2014 after the research card
 
-**RENDER NEXT STEPS via \`ask_user_input_v0\` when the host exposes it.**
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
 
-The (Observation, Suggest, Calls) table below is the source of truth for which moves are valid. Pick the 2\u20134 most relevant rows based on what the response actually contains, then surface them as a \`single_select\` quick-select widget:
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
 
-\`\`\`
-ask_user_input_v0({
-  questions: [{
-    question: "What next?",
-    type: "single_select",
-    options: [
-      "<Suggest column from row 1>",
-      "<Suggest column from row 2>",
-      "<Suggest column from row 3>"
-    ]
-  }]
-})
-\`\`\`
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
 
-When the user picks an option, you call the matching tool from the \`Calls\` column. Constraints carried over from the widget contract: 2\u20134 mutually-exclusive options per question, button-sized labels (\u22646 words), max 3 questions per call.
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
 
-**Fallback prose mode** \u2014 when the host doesn't expose \`ask_user_input_v0\` (or it returned an error): surface the same 2\u20133 picks as a short bulleted list of "Suggest" phrasings. The table itself stays internal; never recite the whole table to the user.
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
 
 ---
 
@@ -8437,6 +8318,7 @@ out?"\`
 
 | Observation                                            | Suggest                                                  | Calls                                                          |
 |--------------------------------------------------------|----------------------------------------------------------|----------------------------------------------------------------|
+| Lead is clearly not a fit (wrong industry, too small)  | "Dislike this lead"                                      | leadbay_dislike_lead({ leadId })                               |
 | User is done with this lead                            | "Back to the inbox"                                      | leadbay_pull_leads                                             |
 
 
@@ -8656,7 +8538,7 @@ Modern chat hosts (Claude, ChatGPT) expose first-party widgets the agent can rou
 |---|---|---|
 | \`places_map_display_v0\` (Claude) | Result has \u22652 leads with \`location.city\` set, and the user's intent is geographic / "in person" / travel | \`{name: lead.company_name, address: "<city>, <country>", place_id: lead.location.place_id ?? omit, notes: <one-sentence pitch>}\` per location |
 | \`message_compose_v1\` (Claude) | You're about to draft outreach (email / message / call opener) | \`{kind: "email", summary_title, variants: [{label, body, subject}]}\` \u2014 2\u20133 variants, labels describe STRATEGY ("Push for alignment", "Reference the M&A signal"), not tone ("Friendly", "Formal") |
-| \`ask_user_input_v0\` (Claude) | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | \`{questions: [{question: "What next?", type: "single_select", options: [<2-4 short button labels>]}]}\`; max 3 questions per call |
+| \`ask_user_input_v0\` (Claude chat / ChatGPT) **or** \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 whichever is in your tool set; their schemas differ, match the one you have | The tool's NEXT STEPS block has 2\u20134 mutually-exclusive next moves and the user hasn't already chosen | Per-tool schema in the server instructions + NEXT STEPS routing block. Max 3 questions. |
 
 ChatGPT exposes the same routing pattern via \`_meta.openai/outputTemplate\`. We don't ship any custom widgets ourselves \u2014 this gate is exclusively about routing into the host's first-party widgets when the data shape fits.
 
@@ -8674,6 +8556,8 @@ WHEN TO USE: the user signals a *mixed* tour-planning intent \u2014 they want bo
 WHEN NOT TO USE: if the user only wants follow-ups (use \`leadbay_followups_map\`), only wants new leads (use \`leadbay_pull_leads\`), wants research on one specific account (\`leadbay_research_lead_by_id\`), or wants to persist the tour as a campaign artifact (chain into \`leadbay_create_campaign\` after this).
 
 **Response envelope**: \`{city, city_id, monitor_leads, discover_leads, discover_filter_note, _meta}\` on happy path; \`{status: "ambiguous_locations", location_ambiguities, ...}\` when the passed \`city\` matched multiple admin areas.
+
+---
 `;
     leadbay_update_lens = `Update lens metadata (name, description, mode flags). Does NOT change the audience filter \u2014 use leadbay_update_lens_filter for that.
 
@@ -13316,6 +13200,35 @@ function summarise(responses) {
   }
   return { answered, total, avg_qualification_boost: avg, best_response_excerpt: excerpt };
 }
+function buildPullLeadsNextSteps(args) {
+  const { leadCount, hasMore, nextPage } = args;
+  if (leadCount <= 0)
+    return null;
+  const options = [];
+  options.push({
+    label: "Triage board",
+    description: "Build an interactive lead triage board to sort and filter this batch.",
+    kind: "build_artifact"
+  });
+  options.push({
+    label: "Deepen qualification",
+    description: "Run deeper AI qualification on these leads.",
+    kind: "qualify_deeper"
+  });
+  if (hasMore && nextPage != null) {
+    options.push({
+      label: "Next page",
+      description: `Pull page ${nextPage + 1} of this lens.`,
+      kind: "pull_next_page"
+    });
+  }
+  options.push({
+    label: "Refine audience",
+    description: "Adjust the lens audience / filters (sector, size, prompt).",
+    kind: "refine_audience"
+  });
+  return { question: "What do you want to do next?", options: options.slice(0, 4) };
+}
 var pullLeads;
 var init_pull_leads = __esm({
   "../core/dist/composite/pull-leads.js"() {
@@ -13386,6 +13299,24 @@ var init_pull_leads = __esm({
             type: "boolean",
             description: "True if scoring is still running."
           },
+          next_steps: {
+            type: ["object", "null"],
+            description: "Ready-made NEXT STEPS for the host's choice widget. Each option has a SHORT `label` (\u22645 words, fits AskUserQuestion's label cap on Claude cowork/Claude Code) and a full `description`. For AskUserQuestion (cowork/Claude Code) pass each option as {label, description}. For ask_user_input_v0 (Claude chat/ChatGPT, string-only options) use the `description` as the option string. Use these VERBATIM, in order \u2014 do NOT re-derive, reword, or render as prose when a widget tool exists. options[0] is the artifact offer (build the lead triage board) whenever the batch is non-empty. null only when the batch is empty.",
+            properties: {
+              question: { type: "string" },
+              options: {
+                type: "array",
+                items: {
+                  type: "object",
+                  properties: {
+                    label: { type: "string" },
+                    description: { type: "string" },
+                    kind: { type: "string" }
+                  }
+                }
+              }
+            }
+          },
           _meta: {
             type: "object",
             description: "Operator context: region + last-call latency.",
@@ -13460,6 +13391,8 @@ var init_pull_leads = __esm({
         const currentPage = res.pagination?.page ?? page;
         const hasMore = currentPage < totalPages - 1;
         const nextPage = hasMore ? currentPage + 1 : null;
+        const leadCount = res.items.length;
+        const nextSteps = buildPullLeadsNextSteps({ leadCount, hasMore, nextPage });
         return withAgentMemoryMeta(client, {
           lens: { id: lensId },
           leads: res.items.map((lead) => ({
@@ -13471,6 +13404,7 @@ var init_pull_leads = __esm({
           next_page: nextPage,
           computing_wishlist: res.computing_wishlist,
           computing_scores: res.computing_scores,
+          next_steps: nextSteps,
           _meta: {
             region: client.region,
             latency_ms: client.lastMeta?.latency_ms ?? null
@@ -20785,10 +20719,10 @@ import {
 var leadbay_daily_check_in = `
 ## MEMORY
 
-Before responding, glance at any \`_meta.agent_memory.summary\` returned by tool calls earlier in this session and reflect its top signals in your reasoning ("Filtering by your stated preference for healthcare"). After any material new taste signal from the user this conversation (sector, region, deal size, communication style, qualification rule, explicit retraction), call \`leadbay_agent_memory_capture\` to persist it: \`source:"user_stated"\` if literal, \`source:"inferred"\` with confidence <=6 if inferred.
+Before responding, glance at any \`_meta.agent_memory.summary\` returned by tool calls earlier in this session and reflect its top signals in your reasoning ("Filtering by your stated preference for healthcare"). After any material new signal from the user this conversation (sector, region, deal size, communication style, qualification rule, explicit retraction, or recurrence / scheduling preference such as "I do this every day" or "remind me every morning"), call \`leadbay_agent_memory_capture\` to persist it: \`source:"user_stated"\` if literal, \`source:"inferred"\` with confidence <=6 if inferred.
 
 
-Run the Leadbay daily check-in for me. Treat this prompt the same way for any equivalent ask focused on NEW leads from the Discover wishlist: "get me leadbay leads", "best NEW leads to prospect today", "what's new today", "show me my batch", "let's prospect". For follow-up phrasings ("what should I follow up on", "leads I've already worked", "before my trip"), this is the wrong prompt \u2014 route to \`leadbay_followup_check_in\` instead. If the user's intent is ambiguous ("what should I work on?"), ASK once before picking an entry point.
+Run the Leadbay daily check-in for me. Treat this prompt the same way for any equivalent ask focused on NEW leads from the Discover wishlist: "get me leadbay leads", "best NEW leads to prospect today", "what's new today", "show me my batch", "let's prospect", "run my morning check-in", "my daily routine", "I do this every day", "every morning". For follow-up phrasings ("what should I follow up on", "leads I've already worked", "before my trip"), this is the wrong prompt \u2014 route to \`leadbay_followup_check_in\` instead. **Recurrence language ("I do this every day", "every morning", "my routine") always means this prompt \u2014 it is a daily batch check-in, not a follow-up.** If the user's intent is ambiguous ("what should I work on?"), ASK once before picking an entry point.
 
 # Resilience rules for Leadbay long-running tools
 
@@ -20914,13 +20848,66 @@ If the batch returns fewer than 10 qualified leads, top it up: call \`leadbay_bu
 
 # PHASE 4 \u2014 DEEP DIVE (every promising lead)
 
+**Skip Phase 4 if the user's request is primarily to view the batch** (e.g., "show me today's leads", "run my morning check-in", "what's in my inbox") \u2014 proceed directly to NEXT STEPS. Run Phase 4 when the user explicitly asks to research leads, names a specific company, or says "and then research" / "deep dive" / "tell me more about".
+
 Call \`leadbay_research_lead_by_id\` on **every** lead from your top 10 that the user might realistically prospect today (filter out clearly weak fits if any). Don't pick just one. **Call it sequentially** \u2014 one at a time, or batches of at most 3 in parallel. Do not fire 10 in parallel \u2014 it triggers transport backpressure that surfaces as \`"Tool permission stream closed"\` errors (see Rule 3 above). If a call fails, retry that single lead once; if the retry also fails, note the lead id and continue. Report Phase 4 results even if 1\u20132 leads were unresearchable.
 
 For each researched lead surface:
 - what makes it promising (1\u20132 sentences citing signals from the research)
 - the **recommended contacts** the research returns \u2014 name, title, why they're the right starting point
 
-Then ASK the user (don't auto-run): "Want me to enrich the contacts on these leads to acquire their emails / phone numbers?" If the user says yes, call \`leadbay_enrich_contacts\` for the relevant lead IDs (this consumes enrichment quota \u2014 that's why we ask first).
+Contact enrichment is offered in the NEXT STEPS widget below \u2014 do NOT emit a separate prose question here. The widget handles the enrichment offer as one of the selectable options. If the user selects enrichment, call \`leadbay_enrich_contacts({leadId, contactId})\` ONCE PER CONTACT \u2014 the tool takes a single leadId + contactId, never a list. (For bulk title/seniority enrichment across many leads at once, use \`leadbay_enrich_titles({leadIds: [...]})\` instead.) This consumes enrichment quota.
+
+# NEXT STEPS
+
+**Sequential request gate:** If the user's original message contained the literal phrase "and then" (e.g., "show me X and then do Y"), and all stated actions have been completed this turn, skip the NEXT STEPS widget entirely and emit STOP directly. The user stated their full plan; they do not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, go directly to STOP without any widget.
+- Do NOT skip for: plain single-action requests ("show me today's leads"), recurring requests ("I do this every day"), or multi-step workflows the user didn't pre-specify.
+
+**REQUIRED OPTIONS \u2014 triggers and position rules:**
+- **Recurring language** ("every day", "every morning", "I do this every", "remind me", "automate this", "recurring"): add "Schedule 'Daily prospecting check-in' as a recurring task" and place it **first**.
+- **\u22655 leads returned**: add "Build an interactive lead triage board for this batch" and place it **first** (or second if the scheduling offer above also applies).
+
+## NEXT STEPS \u2014 after rendering the pull_leads table
+
+**ALWAYS render NEXT STEPS via your host's next-step widget.** Use whichever is in your tool set \u2014 the NAME and SCHEMA differ: **\`ask_user_input_v0\`** (Claude chat / ChatGPT) takes plain-string options with \`type:"single_select"\`; **\`AskUserQuestion\`** (Claude cowork / Claude Code) takes object options \`{label, description}\` plus a required short \`header\` (\u226412 chars) and \`multiSelect\`, NO \`type\` field, and never add an "Other" option (the host adds it). Match the schema to the tool you actually have \u2014 the wrong schema fails silently and you fall back to prose. Prose bullets are the fallback ONLY when NEITHER widget exists. Any turn that would end with a choice must be the widget \u2014 the widget IS the question.
+
+**If the tool result carries a \`next_steps\` object, that is the source of truth \u2014 use it directly.** Each option has a short \`.label\` (\u22645 words) and a full \`.description\`. Map \`next_steps.options[]\` into your host widget VERBATIM and in order: for \`AskUserQuestion\` (cowork / Claude Code) pass each as \`{label, description}\`; for \`ask_user_input_v0\` (Claude chat / ChatGPT, string options only) pass each option's \`.description\` as the string (it's the full sentence). Do NOT reword, reorder, drop, or prose-ify them \u2014 they're built deterministically by the server so the offer (incl. the artifact option at position 0) fires every time. Fall back to the table below only when there is NO \`next_steps\` field.
+
+**One exception \u2014 skip the widget** when the user's original message contained a complete sequential instruction chain ("show me X and then do Y") AND all stated steps have been completed. In that case, end with STOP directly \u2014 the user stated their full plan and does not need a "what next?" prompt.
+- Skip example: "Show me today's leads and then research the top one for me." \u2192 after research completes, emit STOP without the widget.
+- Do NOT skip for: plain requests ("show me today's leads", "run my check-in"), recurring-language requests ("I do this every day"), or requests where only one action was stated.
+
+Pick 2\u20134 rows from the (Observation, Suggest, Calls) table below most relevant to the response, then call your host's widget with ITS schema (per the schema rules above \u2014 wrong schema fails silently):
+- \`ask_user_input_v0\`: \`{questions:[{question,type:"single_select",options:["<Suggest 1>","<Suggest 2>"]}]}\`
+- \`AskUserQuestion\`: \`{questions:[{question,header:"Next step",multiSelect:false,options:[{label:"<\u22645 words>",description:"<Suggest 1>"}]}]}\`
+
+User picks \u2192 call the matching \`Calls\` tool. Constraints: 2\u20134 mutually-exclusive options, AskUserQuestion labels \u22645 words (full text in \`description\`), max 3 questions. Table stays internal; never recite it.
+
+---
+
+
+
+Pick 2\u20133 items below based on what was actually observed in the response. The table is the source of truth for which moves are valid.
+
+| Observation                                                | Suggest                                                      | Calls                                                  |
+|------------------------------------------------------------|--------------------------------------------------------------|--------------------------------------------------------|
+| \u2265 5 leads returned (any batch)                             | "Build an interactive lead triage board for this batch"      | emit antArtifact from data in hand (do NOT re-call leadbay_pull_leads) |
+| \`has_more == true\`                                         | "Pull the next page (page N+1 of M)"                         | leadbay_pull_leads(page = current + 1, lensId = pinned)|
+| \u2265 3 rows have \`qualification_summary.answered == 0\`        | "Deepen AI qualification on the rows without \u2756 caps"         | leadbay_bulk_qualify_leads(leadIds=[\u2026])                |
+| User points at a single row                                | "Research [Company] in depth"                                | leadbay_research_lead_by_id(leadId)                    |
+| User only has a name (no leadId in context)                | "Look up [Company] by name"                                  | leadbay_research_lead_by_name_fuzzy(companyName)       |
+| Top row has phone AND email                                | "Prepare an outreach for [Contact] \u2014 call + email"           | leadbay_prepare_outreach(leadId)                       |
+| Top row has email but no phone                             | "Draft an outreach email for [Contact]"                      | leadbay_prepare_outreach(leadId)                       |
+| Top row has phone but no email                             | "Show [Contact]'s call details + a 60-second opener"         | leadbay_prepare_outreach(leadId)                       |
+| Top row has contacts but no phone/email                    | "Order contact enrichment to surface email/phone first"      | leadbay_enrich_titles(...) or leadbay_prepare_outreach(leadId, enrich:true) |
+| \`computing_scores == true\` or \`computing_wishlist == true\` | "Scores are still being computed \u2014 re-pull in ~30s"          | leadbay_pull_leads (retry with same lensId)            |
+| User wants a narrower / wider audience                     | "Adjust the lens filters (sector / size)"                    | leadbay_adjust_audience(...)                           |
+| Phase 4 research was run (\`research_lead_by_id\` called) AND top contacts lack direct email/phone | "Enrich contacts on [Lead1], [Lead2] to get direct emails and phone numbers" | leadbay_enrich_contacts(leadId, contactId) \u2014 ONE call per contact (the tool takes a single leadId + contactId, never a list) |
+If nothing in the menu applies cleanly, suggest only "pull next page" and "research a specific lead in depth" \u2014 never invent a tool that doesn't exist.
+
+
+**Final ordering check (do this before rendering):** Recurring offer \u2192 option 1; triage board \u2192 option 1 (or 2 if scheduling is also required). Swap if needed.
 
 # GATE \u2014 STOP
 
@@ -21167,7 +21154,7 @@ Done. The map is the surface; the drafts are the action; the campaign is the per
 var leadbay_prospecting_overview = `
 ## MEMORY
 
-Before responding, glance at any \`_meta.agent_memory.summary\` returned by tool calls earlier in this session and reflect its top signals in your reasoning ("Filtering by your stated preference for healthcare"). After any material new taste signal from the user this conversation (sector, region, deal size, communication style, qualification rule, explicit retraction), call \`leadbay_agent_memory_capture\` to persist it: \`source:"user_stated"\` if literal, \`source:"inferred"\` with confidence <=6 if inferred.
+Before responding, glance at any \`_meta.agent_memory.summary\` returned by tool calls earlier in this session and reflect its top signals in your reasoning ("Filtering by your stated preference for healthcare"). After any material new signal from the user this conversation (sector, region, deal size, communication style, qualification rule, explicit retraction, or recurrence / scheduling preference such as "I do this every day" or "remind me every morning"), call \`leadbay_agent_memory_capture\` to persist it: \`source:"user_stated"\` if literal, \`source:"inferred"\` with confidence <=6 if inferred.
 
 
 # Leadbay Prospecting \u2014 Orientation
@@ -21441,7 +21428,7 @@ If the response status is \`applied\`, tell me Leadbay is regenerating intellige
 var leadbay_research_a_domain = `
 ## MEMORY
 
-Before responding, glance at any \`_meta.agent_memory.summary\` returned by tool calls earlier in this session and reflect its top signals in your reasoning ("Filtering by your stated preference for healthcare"). After any material new taste signal from the user this conversation (sector, region, deal size, communication style, qualification rule, explicit retraction), call \`leadbay_agent_memory_capture\` to persist it: \`source:"user_stated"\` if literal, \`source:"inferred"\` with confidence <=6 if inferred.
+Before responding, glance at any \`_meta.agent_memory.summary\` returned by tool calls earlier in this session and reflect its top signals in your reasoning ("Filtering by your stated preference for healthcare"). After any material new signal from the user this conversation (sector, region, deal size, communication style, qualification rule, explicit retraction, or recurrence / scheduling preference such as "I do this every day" or "remind me every morning"), call \`leadbay_agent_memory_capture\` to persist it: \`source:"user_stated"\` if literal, \`source:"inferred"\` with confidence <=6 if inferred.
 
 
 IRON LAW \u2014 NO FABRICATION. Every lead id, contact email, custom field id, mapping decision, and tool argument must trace to a value you read from the file the user attached or to an output from a leadbay_* tool call in this session. Do not invent values. Do not "fill in" a missing leadId with a name match. Do not synthesize a CRM id from a guess. If a value is missing, leave the field blank and say so.
@@ -21582,7 +21569,7 @@ Then ask me ONCE: "Which of these should we drop?" If I name leads to drop, excl
 
 # PHASE 3 \u2014 DECIDE THE CAMPAIGN SHAPE
 
-If I provided a \`rep_split\` ("one campaign per rep: John gets Tulsa, Sarah gets OKC"), partition the validated leads accordingly. If I didn't, ask ONCE: "Create one campaign for the whole batch, or split per rep / region / sector?" \u2014 surface 2-4 options via \`ask_user_input_v0\` when available, else as a bulleted list.
+If I provided a \`rep_split\` ("one campaign per rep: John gets Tulsa, Sarah gets OKC"), partition the validated leads accordingly. If I didn't, ask ONCE: "Create one campaign for the whole batch, or split per rep / region / sector?" \u2014 surface 2-4 options via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`) when available, else as a bulleted list.
 
 For each campaign-shape decision, derive a name. Templates:
 - Whole batch: \`"<lens-name> \u2013 <YYYY-MM-DD>"\`
@@ -21599,7 +21586,7 @@ Once the campaigns are created, surface this caveat in plain prose:
 
 > Campaign visibility is currently scoped to the user who CREATED the campaign \u2014 the reps won't see these in their own MCP \`leadbay_list_campaigns\` calls. They CAN see them in the web UI at app.leadbay.ai \u2192 Campaigns. Cross-user MCP visibility would need backend work; flag this as a #3630 US3 product gap if your reps work primarily through MCP.
 
-End with a NEXT STEPS chip via \`ask_user_input_v0\`: "View progression on one of these now?" \u2192 routes to \`leadbay_campaign_progression\`.
+End with a NEXT STEPS chip via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`): "View progression on one of these now?" \u2192 routes to \`leadbay_campaign_progression\`.
 
 # PHASE 6 \u2014 STOP
 
@@ -21615,7 +21602,7 @@ If the prompt's body and the tool's RENDERING appear to conflict, the tool's REN
 
 # PHASE 0 \u2014 PICK THE CAMPAIGN
 
-If I gave you a name or id, resolve it. Otherwise call \`leadbay_list_campaigns()\` and surface the active campaigns as a \`single_select\` via \`ask_user_input_v0\` (cap at 4 \u2014 sort by \`updated_at\` desc, archived hidden):
+If I gave you a name or id, resolve it. Otherwise call \`leadbay_list_campaigns()\` and surface the active campaigns as a \`single_select\` via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`) (cap at 4 \u2014 sort by \`updated_at\` desc, archived hidden):
 
 > Which campaign do you want to work?
 > - <Name 1> \xB7 <N leads> \xB7 last touched <date>
@@ -21648,7 +21635,7 @@ Call \`leadbay_campaign_call_sheet({campaign_id})\`. The response carries \`summ
 \u{1F4CB} <total_leads> leads \xB7 \u{1F4DE} <leads_with_phone> with a phone \xB7 \u2709 <leads_with_email> with an email \xB7 \u{1F5FA} <leads_with_coords> with coords \xB7 \u{1F534} <leads_without_contacts> need enrichment \xB7 \u2705 <leads_already_contacted> already touched
 \`\`\`
 
-**Then PROPOSE the right modes via \`ask_user_input_v0\`** (2-4 options, sorted by what makes the most sense for THIS campaign's data):
+**Then PROPOSE the right modes via your host's choice widget (\`ask_user_input_v0\` or \`AskUserQuestion\`)** (2-4 options, sorted by what makes the most sense for THIS campaign's data):
 
 - "\u{1F4DE} Start calling now" \u2014 IF \`ready_for_calling\`. Top option when phones are there.
 - "\u2709 Email session instead" \u2014 IF \`ready_for_emailing\` AND \`email_ratio > phone_ratio\`. Don't surface this when calling is more obvious.
@@ -21710,23 +21697,9 @@ Optional: offer to review the \`leadbay_campaign_progression\` for the same camp
 - If the user dictates an outcome that doesn't cleanly map to one of the four epilogue values, ASK ONCE before guessing.
 `;
 var PROMPT_META = {
-  leadbay_daily_check_in: { "name": "leadbay_daily_check_in", "short_description": `Run the canonical daily check-in: account state, fresh batch, triage
-top 10, deep-dive every promising one, offer contact enrichment. The
-morning DISCOVERY workflow (new leads from the lens wishlist). Trigger
-on "leadbay leads", "best NEW leads", "what's new today", "show me the
-day's batch", "let's prospect". Do NOT trigger on follow-up phrasings
-("what should I follow up on", "before my trip") \u2014 those go to
-\`leadbay_followup_check_in\`.
-`, "arguments": [], "expected_calls": ["leadbay_account_status", "leadbay_pull_leads", "leadbay_research_lead_by_id", "leadbay_bulk_qualify_leads", "leadbay_enrich_contacts"], "failure_modes": ["Calls leadbay_report_outreach without explicit user authorization", "Surfaces fewer than 10 leads when more are available, or fails to top up via leadbay_qualify_top_n when the batch is short", `Replaces the canonical pull_leads table layout with prose per row (the per-tool RENDERING block is the structural contract; "Today's nudges" goes above it, not in place of it)`, "Skips the nudge paragraph entirely \u2014 the table alone is fine but adding the nudge is the value-add", "Skips deep research on promising leads (Phase 4) \u2014 the agent must call leadbay_research_lead_by_id on each, not just one", "Triggers contact enrichment without asking the user first (it consumes quota)", "Skips the STOP byproduct and proposes next actions on its own", 'Fires 10 parallel leadbay_research_lead_by_id calls and treats "stream closed" errors as terminal \u2014 must serialize and retry singletons', "Re-pulls leadbay_pull_leads without passing the captured lensId, allowing a backend lens shift to discard the Phase 2 batch", 'Treats a "Request timed out" from leadbay_bulk_qualify_leads as terminal instead of retrying with wait_for_completion:false + qualify_status polling', 'Triggers on a follow-up query (e.g., "leads I should follow up with") that should have routed to `leadbay_followup_check_in` \u2014 the two entry points are different data sources (Discover wishlist vs Monitor view) per \xA71.6'] },
-  leadbay_extend_my_lens: { "name": "leadbay_extend_my_lens", "short_description": "Add more leads to the current lens on demand \u2014 for users whose appetite\nexceeds the standard daily fill. The agent picks seeds silently from\nwhat's already on the lens, fires the extra refill, and surfaces the\nqueue confirmation. The user never reviews the seed list.\n", "arguments": [{ "name": "extra_count", "description": "How many extra leads to add. Optional. Omit to use the backend default.", "required": false }], "expected_calls": ["leadbay_account_status", "leadbay_seed_candidates", "leadbay_extend_lens", "leadbay_pull_leads"], "failure_modes": ["Surfaces the seed candidate list to the user instead of picking silently \u2014 the user asked for MORE LEADS, not a candidate review meeting", "Skips the seeded path and calls `leadbay_extend_lens` with no `seed_lead_ids`, losing the bias signal the recommender needs", "On 429, silently retries instead of surfacing the three options (smaller / wait / upgrade) via `ask_user_input_v0`", "Forgets to pre-check `LENS_EXTRA_REFILL` quota in `leadbay_account_status` and burns a wasted API call", "Skips the post-queue pull-leads suggestion, so the user doesn't see what just got added"] },
-  leadbay_followup_check_in: { "name": "leadbay_followup_check_in", "short_description": `Run the canonical follow-up check-in: surface KNOWN leads from the
-Monitor view that need re-engagement today, ranked by AI urgency,
-with the canonical pull_followups table layout. Trigger when the
-user asks "follow up", "already known leads", "leads I haven't
-contacted", "leads in [city]", "before my trip", "this week",
-"this month", "what's overdue", "who should I re-engage", or
-anything that implies pre-existing pipeline context.
-`, "arguments": [], "expected_calls": ["leadbay_pull_followups", "leadbay_research_lead_by_id", "leadbay_prepare_outreach"], "failure_modes": ["Calls leadbay_pull_leads (the Discover entry point) instead of leadbay_pull_followups \u2014 these are different data sources; the Discover queue does NOT contain Monitor's known-but-cold pipeline", 'Iterates pages of leadbay_pull_leads filtering by engagement_count to "fake" a follow-up view (a real bug observed in 0.9.0 \u2014 the right move is to call pull_followups directly)', "Replaces the canonical pull_followups table layout with prose per row (the per-tool RENDERING block is the structural contract; commentary belongs above or below)", 'Skips the cross-mode pivot offer at the end ("Want to see NEW leads from your wishlist instead?" routes to leadbay_pull_leads)'] },
+  leadbay_daily_check_in: { "name": "leadbay_daily_check_in", "short_description": 'Morning DISCOVERY workflow \u2014 new leads from the lens wishlist. Trigger\non "show me leads", "what\'s new today", "let\'s prospect", "run my check-in",\n"my morning check-in", "I do this every day", "every morning". Recurrence\nlanguage always means this prompt. Do NOT trigger on follow-up phrasings\n("follow up", "before my trip") \u2014 those go to `leadbay_followup_check_in`.\n', "arguments": [], "expected_calls": ["leadbay_account_status", "leadbay_pull_leads", "leadbay_research_lead_by_id", "leadbay_bulk_qualify_leads", "leadbay_enrich_contacts"], "failure_modes": ["Calls leadbay_report_outreach without explicit user authorization", "Surfaces fewer than 10 leads when more are available, or fails to top up via leadbay_qualify_top_n when the batch is short", `Replaces the canonical pull_leads table layout with prose per row (the per-tool RENDERING block is the structural contract; "Today's nudges" goes above it, not in place of it)`, "Skips the nudge paragraph entirely \u2014 the table alone is fine but adding the nudge is the value-add", `Skips deep research on promising leads (Phase 4) \u2014 the agent must call leadbay_research_lead_by_id on each when the user's intent is to research specific leads; Phase 4 is intentionally skipped for batch-view requests ("show me today's leads", "run my morning check-in") per the Phase 4 skip gate`, "Triggers contact enrichment without asking the user first (it consumes quota)", "Skips the STOP byproduct and proposes next actions on its own", 'Fires 10 parallel leadbay_research_lead_by_id calls and treats "stream closed" errors as terminal \u2014 must serialize and retry singletons', "Re-pulls leadbay_pull_leads without passing the captured lensId, allowing a backend lens shift to discard the Phase 2 batch", 'Treats a "Request timed out" from leadbay_bulk_qualify_leads as terminal instead of retrying with wait_for_completion:false + qualify_status polling', 'Triggers on a follow-up query (e.g., "leads I should follow up with") that should have routed to `leadbay_followup_check_in` \u2014 the two entry points are different data sources (Discover wishlist vs Monitor view) per \xA71.6'] },
+  leadbay_extend_my_lens: { "name": "leadbay_extend_my_lens", "short_description": "Add more leads to the current lens on demand \u2014 for users whose appetite\nexceeds the standard daily fill. The agent picks seeds silently from\nwhat's already on the lens, fires the extra refill, and surfaces the\nqueue confirmation. The user never reviews the seed list.\n", "arguments": [{ "name": "extra_count", "description": "How many extra leads to add. Optional. Omit to use the backend default.", "required": false }], "expected_calls": ["leadbay_account_status", "leadbay_seed_candidates", "leadbay_extend_lens", "leadbay_pull_leads"], "failure_modes": ["Surfaces the seed candidate list to the user instead of picking silently \u2014 the user asked for MORE LEADS, not a candidate review meeting", "Skips the seeded path and calls `leadbay_extend_lens` with no `seed_lead_ids`, losing the bias signal the recommender needs", "On 429, silently retries instead of surfacing the three options (smaller / wait / upgrade) via your host's choice widget (`ask_user_input_v0` or `AskUserQuestion`)", "Forgets to pre-check `LENS_EXTRA_REFILL` quota in `leadbay_account_status` and burns a wasted API call", "Skips the post-queue pull-leads suggestion, so the user doesn't see what just got added"] },
+  leadbay_followup_check_in: { "name": "leadbay_followup_check_in", "short_description": 'Follow-up check-in: surface KNOWN leads from the Monitor view needing\nre-engagement. Trigger on "follow up", "already known leads", "what\'s\noverdue", "before my trip", "who should I re-engage". Do NOT trigger on\n"show me today\'s leads", "my morning check-in", "run my check-in",\n"I do this every day", "every morning" \u2014 those go to\n`leadbay_daily_check_in`.\n', "arguments": [], "expected_calls": ["leadbay_pull_followups", "leadbay_research_lead_by_id", "leadbay_prepare_outreach"], "failure_modes": ["Calls leadbay_pull_leads (the Discover entry point) instead of leadbay_pull_followups \u2014 these are different data sources; the Discover queue does NOT contain Monitor's known-but-cold pipeline", 'Iterates pages of leadbay_pull_leads filtering by engagement_count to "fake" a follow-up view (a real bug observed in 0.9.0 \u2014 the right move is to call pull_followups directly)', "Replaces the canonical pull_followups table layout with prose per row (the per-tool RENDERING block is the structural contract; commentary belongs above or below)", 'Skips the cross-mode pivot offer at the end ("Want to see NEW leads from your wishlist instead?" routes to leadbay_pull_leads)'] },
   leadbay_import_file: { "name": "leadbay_import_file", "short_description": "Import a user-supplied CSV/file into Leadbay through five phases with\nevidence gates \u2014 scan, derive, resolve identities, preserve & commit,\nthen optionally qualify and report. The job is to maximize how many\nrows the Leadbay system actually ingests and matches.\n", "arguments": [{ "name": "file", "description": "Path or user-visible name of the CSV/file to import. If omitted, use the file the user attached or referenced.", "required": false }, { "name": "instruction", "description": 'Additional user goal, e.g. "then qualify the leads", "preserve owner phone as a custom field", or "only import restaurants in Manhattan".', "required": false }], "expected_calls": ["leadbay_resolve_import_rows", "leadbay_list_mappable_fields", "leadbay_create_custom_field", "leadbay_import_leads", "leadbay_import_and_qualify", "leadbay_add_note", "leadbay_import_status"], "failure_modes": ["Picks LEADBAY_ID from score alone, name-only, fuzzy-name-only, root-domain-only, brand-only, postcode-only, or city-only evidence", "Drops meaningful business notes or CRM record links instead of preserving them as custom fields or lead notes", "Treats a consumer mailbox domain (gmail.com, hotmail.com, ...) as the company domain", "Skips deriving company_domain from a business email when no website column exists (this kills match rate)", "Skips the COLUMN PRESERVATION PLAN byproduct before importing", "Skips the DECISION LOG byproduct before writing LEADBAY_ID", "Returns the imported records WITHOUT writing LEADBAY_ID values back into the user's file (leaves the user no audit trail of what matched)", "Fabricates leadIds, contact emails, or mapping IDs not present in the file or a tool response"] },
   leadbay_log_outreach: { "name": "leadbay_log_outreach", "short_description": "Log outreach (an email I sent, a call I made, a meeting I had) on a\nspecific lead. Captures verification so the SDR pipeline trusts the entry.\n", "arguments": [{ "name": "lead_id", "description": "The lead UUID. Get it from leadbay_pull_leads or leadbay_research_lead_by_id.", "required": true }, { "name": "summary", "description": "1-2 sentences describing what I did (e.g. 'Sent intro email to CTO citing recent Hornsea contract').", "required": true }], "expected_calls": ["leadbay_report_outreach"], "failure_modes": ["Calls leadbay_report_outreach without first collecting a verification source", "Fabricates a gmail_message_id or calendar_event_id (the human team treats verification as canonical)", "Records outreach to a different lead_id than the one the user supplied", "Skips the dry_run step when the user is unsure what would be sent"] },
   leadbay_plan_tour_in_city: { "name": "leadbay_plan_tour_in_city", "short_description": "Plan a field sales tour: in one flow, surface follow-ups + fresh\nDiscover leads in the target city via `leadbay_tour_plan`, render\nto a map, draft in-area outreach via `leadbay_prepare_outreach`,\nand optionally persist the selected accounts as a named campaign\nvia `leadbay_create_campaign`. Closes #3630 US1 end-to-end.\n", "arguments": [{ "name": "city", "description": "City or region the user is visiting (e.g. 'Limoges', 'Bay Area'). Used as the geo filter for both Monitor and Discover lookups.", "required": true }, { "name": "date", "description": "When the visit is (e.g. 'May 24', 'next Thursday'). Surfaced in the outreach drafts as 'I'll be in <city> on <date>'.", "required": false }], "expected_calls": ["leadbay_tour_plan", "leadbay_research_lead_by_id", "leadbay_prepare_outreach", "leadbay_create_campaign"], "failure_modes": ["Calls leadbay_followups_map (Monitor-only) instead of leadbay_tour_plan \u2014 loses the Discover (fresh-lead) half that the user explicitly asked for", "Calls leadbay_pull_leads then drops the geo filter \u2014 returns the lens-wide wishlist instead of city-relevant fresh leads", 'Skips the campaign-persist step ("would you like to save these as a tour?") \u2014 leaves the rep with a one-shot map but no follow-up artifact', "Creates a campaign WITHOUT asking the user first \u2014 the persist step is high-intent; offer it, don't assume", "Fabricates lead_ids when seeding the campaign instead of using the ids returned by tour_plan"] },
@@ -21745,9 +21718,9 @@ should I follow up on" to "I'll send via lemlist".
 };
 var PROMPT_CATALOG_HEADER = `This server exposes the following workflow prompts via \`prompts/list\` and \`prompts/get\`. Some MCP clients render them as slash commands; if your client does not, you (the agent) should invoke them directly via \`prompts/get\` when the user's request matches one of the triggers described below.`;
 var PROMPT_CATALOG_BULLETS = {
-  leadbay_daily_check_in: `- \`leadbay_daily_check_in\`: Run the canonical daily check-in: account state, fresh batch, triage top 10, deep-dive every promising one, offer contact enrichment. The morning DISCOVERY workflow (new leads from the lens wishlist). Trigger on "leadbay leads", "best NEW leads", "what's new today", "show me the day's batch", "let's prospect". Do NOT trigger on follow-up phrasings ("what should I follow up on", "before my trip") \u2014 those go to \`leadbay_followup_check_in\`.`,
+  leadbay_daily_check_in: `- \`leadbay_daily_check_in\`: Morning DISCOVERY workflow \u2014 new leads from the lens wishlist. Trigger on "show me leads", "what's new today", "let's prospect", "run my check-in", "my morning check-in", "I do this every day", "every morning". Recurrence language always means this prompt. Do NOT trigger on follow-up phrasings ("follow up", "before my trip") \u2014 those go to \`leadbay_followup_check_in\`.`,
   leadbay_extend_my_lens: `- \`leadbay_extend_my_lens\` (optional args: extra_count): Add more leads to the current lens on demand \u2014 for users whose appetite exceeds the standard daily fill. The agent picks seeds silently from what's already on the lens, fires the extra refill, and surfaces the queue confirmation. The user never reviews the seed list.`,
-  leadbay_followup_check_in: `- \`leadbay_followup_check_in\`: Run the canonical follow-up check-in: surface KNOWN leads from the Monitor view that need re-engagement today, ranked by AI urgency, with the canonical pull_followups table layout. Trigger when the user asks "follow up", "already known leads", "leads I haven't contacted", "leads in [city]", "before my trip", "this week", "this month", "what's overdue", "who should I re-engage", or anything that implies pre-existing pipeline context.`,
+  leadbay_followup_check_in: `- \`leadbay_followup_check_in\`: Follow-up check-in: surface KNOWN leads from the Monitor view needing re-engagement. Trigger on "follow up", "already known leads", "what's overdue", "before my trip", "who should I re-engage". Do NOT trigger on "show me today's leads", "my morning check-in", "run my check-in", "I do this every day", "every morning" \u2014 those go to \`leadbay_daily_check_in\`.`,
   leadbay_import_file: `- \`leadbay_import_file\` (optional args: file, instruction): Import a user-supplied CSV/file into Leadbay through five phases with evidence gates \u2014 scan, derive, resolve identities, preserve & commit, then optionally qualify and report. The job is to maximize how many rows the Leadbay system actually ingests and matches.`,
   leadbay_log_outreach: `- \`leadbay_log_outreach\` (required args: lead_id, summary): Log outreach (an email I sent, a call I made, a meeting I had) on a specific lead. Captures verification so the SDR pipeline trusts the entry.`,
   leadbay_plan_tour_in_city: `- \`leadbay_plan_tour_in_city\` (required args: city; optional args: date): Plan a field sales tour: in one flow, surface follow-ups + fresh Discover leads in the target city via \`leadbay_tour_plan\`, render to a map, draft in-area outreach via \`leadbay_prepare_outreach\`, and optionally persist the selected accounts as a named campaign via \`leadbay_create_campaign\`. Closes #3630 US1 end-to-end.`,
@@ -22084,7 +22057,7 @@ async function readResource(uri, client) {
 }
 
 // src/host-widgets.ts
-var BUILTIN_WIDGETS_PARAGRAPH = 'Prefer host-native widgets over inline markdown when the data shape fits. Three to know: (1) `places_map_display_v0` \u2014 for \u22652 locations / map / travel intent. Pass `{name, address, latitude, longitude, notes}` per location; the host enriches via Google Places. (2) `message_compose_v1` \u2014 for any outreach draft (email / message / call opener). Pass 2\u20133 strategic variants with goal-oriented labels ("Push for alignment", "Reference M&A signal") \u2014 NOT tone labels. (3) `ask_user_input_v0` \u2014 for the NEXT STEPS questions every Leadbay tool emits. Pass `single_select` with 2\u20134 mutually-exclusive options from the tool\'s NEXT STEPS table. When the host doesn\'t expose the named widget, fall back to the per-tool markdown RENDERING block. The directive is host-conditional; the fallback is automatic.';
+var BUILTIN_WIDGETS_PARAGRAPH = 'Prefer host-native widgets over inline markdown when the data shape fits. Three to know: (1) `places_map_display_v0` \u2014 for \u22652 locations / map / travel intent. Pass `{name, address, latitude, longitude, notes}` per location; the host enriches via Google Places. (2) `message_compose_v1` \u2014 for any outreach draft (email / message / call opener). Pass 2\u20133 strategic variants with goal-oriented labels ("Push for alignment", "Reference M&A signal") \u2014 NOT tone labels. (3) The next-step / choice widget \u2014 for the NEXT STEPS questions every Leadbay tool emits. Its NAME AND SCHEMA differ by host; use whichever is in your tool set: (a) `ask_user_input_v0` (Claude chat / ChatGPT) \u2014 options are PLAIN STRINGS with `type: "single_select"`, e.g. `{questions:[{question:"What next?",type:"single_select",options:["Build a triage board","Pull next page"]}]}`. (b) `AskUserQuestion` (Claude cowork / Claude Code) \u2014 options are OBJECTS `{label, description}`, plus a required short `header` (\u226412 chars) and a `multiSelect` boolean, and NO `type` field; do not add an "Other" option (the host adds it). e.g. `{questions:[{question:"What next?",header:"Next step",multiSelect:false,options:[{label:"Triage board",description:"Build an interactive board to sort this batch."},{label:"Next page",description:"Pull page 2."}]}]}`. Match the schema to the tool you actually have \u2014 using the string-schema for AskUserQuestion (or vice-versa) makes the call fail and you silently fall back to prose. When the host exposes NEITHER widget, fall back to the per-tool markdown RENDERING block.\n\nWIDGET IS MANDATORY WHEN AVAILABLE: if EITHER `ask_user_input_v0` OR `AskUserQuestion` is present in your tool set, you MUST emit your NEXT STEPS / scheduling / artifact offer by CALLING that widget tool (with its correct schema above) \u2014 do NOT write the options out as a prose question ("Want me to run this every morning?", "Should I build a board?"). Prose for these offers is the FALLBACK reserved ONLY for hosts that expose neither widget. When the widget exists, presenting the same choices as prose instead of calling it is a defect: the user loses the click-to-select surface. So whenever you have a recurring-task offer, an artifact offer, or a 2\u20134 option next-step menu AND a widget tool is available \u2192 call it, every time.';
 
 // src/server.ts
 init_dist();
@@ -22730,6 +22703,41 @@ var QUOTA_TOPUP = `Quota & top-ups: when a tool returns QUOTA_EXCEEDED / 429, th
 var VERIFICATION = `After every email, call, message, or meeting with a lead's contact, you MUST call leadbay_report_outreach with verification={source, ref} (gmail_message_id from the Gmail send, calendar_event_id from a booking, or user_confirmed='<the user's literal confirmation>'). Skipping or fabricating verification poisons the human team's pipeline.`;
 
 // src/server.ts
+var ARTIFACT_PROPOSAL_PARAGRAPH = 'Artifact proposals: after any result the user would scan, compare, plan against, or return to \u2014 or any multi-item result \u2014 DEFAULT TO offering an artifact. Treat the offer as expected, not optional: a multi-item lead batch, a side-by-side comparison, or a research record the user will revisit each SHOULD get an artifact offer unless it clearly falls in the Skip list below. When in doubt, offer. WIDGET IS MANDATORY WHEN AVAILABLE: if your host\'s choice widget is in your tool set \u2014 `ask_user_input_v0` (Claude chat / ChatGPT) OR `AskUserQuestion` (Claude cowork / Claude Code) \u2014 you MUST make the offer by CALLING that widget tool \u2014 never as a prose line ("Want me to build an interactive lead triage board?"). Prose is the fallback ONLY for hosts that expose neither widget. Presenting the offer as prose when a widget exists is a defect \u2014 the user loses the click-to-select surface.\n\nTwo tracks:\n1. A NEXT STEPS menu is already firing (the choice widget with 2\u20134 options): include the artifact offer as one of the slots. It counts toward the 2\u20134 budget \u2014 replace the lowest-priority option when already at 4.\n2. No NEXT STEPS menu is firing: fire a standalone yes/no question by CALLING your host\'s choice widget with ITS schema (see the host-widget section: `ask_user_input_v0` takes string options; `AskUserQuestion` takes {label, description} objects with a `header` and `multiSelect`). The question is "Want me to build a [concrete artifact name] for this?" with yes/no options. Use the schema that matches the widget tool you actually have.\n\nName the artifact concretely from the result you just returned: "Build an interactive lead triage board", "Build a research one-pager for Acme Corp", "Build a side-by-side outreach variant comparison". Never use generic labels.\n\nOn selection of yes/build: build the artifact from data already in hand \u2014 do NOT re-call the tool. Emit with the antArtifact tag: type="text/html" for interactive boards/tables, type="application/vnd.ant.react" for React components, type="text/markdown" for one-pagers.\n\nSkip for: single-action confirmations (like/dislike, report_outreach), mid-flow steps, results a host widget already covers (places_map_display_v0, message_compose_v1).';
+var SCHEDULED_TASK_PARAGRAPH = `Scheduled task proposals: check the user's LITERAL words. NEVER fire based on the workflow or prompt type \u2014 only on what the user ACTUALLY TYPED.
+
+NEVER fire for these examples (no recurring language in user message):
+  - 'Show me today\\'s leads.' \u2192 NO
+  - 'What are my leads today?' \u2192 NO
+  - 'Pull my best new prospects.' \u2192 NO
+  - 'Show me my batch.' \u2192 NO
+  - 'Run my morning check-in.' (no recurring qualifier) \u2192 NO
+
+ONLY fire when the user's message contains explicit recurring language:
+  - 'I do this every day' \u2192 YES
+  - 'every morning' \u2192 YES
+  - 'remind me every week' \u2192 YES
+  - 'automate this' \u2192 YES
+  - 'set this up as recurring' \u2192 YES
+
+WIDGET IS MANDATORY WHEN AVAILABLE: if your host's choice widget is in your tool set \u2014 \`ask_user_input_v0\` (Claude chat / ChatGPT) OR \`AskUserQuestion\` (Claude cowork / Claude Code) \u2014 you MUST make the scheduling offer by CALLING that widget tool \u2014 never as a prose line ("Want me to run this every morning?"). Prose is the fallback ONLY for hosts that expose neither widget; presenting the offer as prose when a widget exists is a defect.
+
+Render EVERY question below by CALLING your host's choice widget with ITS schema (see the host-widget section: \`ask_user_input_v0\` takes string options; \`AskUserQuestion\` takes {label, description} objects with a \`header\` and \`multiSelect\`). Use the schema that matches the widget tool you actually have \u2014 never the other one.
+
+Two tracks \u2014 same as artifact proposals:
+1. A NEXT STEPS menu is already firing: add "Schedule [Task Name] as a recurring task" as one slot (counts toward the 2\u20134 budget \u2014 replace the lowest-priority option when already at 4).
+2. No NEXT STEPS menu is firing: fire a standalone yes/no question "Want me to schedule this as a recurring task?" (options: yes / no thanks).
+
+If the user selects yes, continue with a multi-step flow, one widget call per step:
+Step 1 \u2014 frequency: "How often?" \u2192 Every day / Every weekday / Weekly / Custom.
+Step 2a \u2014 if "Every day" or "Every weekday": "What time?" \u2192 Morning (8am) / Midday (12pm) / Afternoon (5pm) / Custom.
+Step 2b \u2014 if "Weekly": "Which day?" \u2192 Monday / Wednesday / Friday / Custom.
+Step 2c \u2014 if "Custom" at any step: ask for a free-text description and interpret it to determine the schedule.
+After the schedule is confirmed: judge whether the scheduled run should also produce an artifact. If yes, offer "Should each run also build an artifact (e.g. a fresh lead board)?" \u2192 yes / no.
+
+Name the task concretely from context: "Daily prospecting check-in", "Weekly follow-up sweep", "Monday morning lead review". Never use generic labels.
+
+Skip for: single-action confirmations, mid-flow steps, one-off lookups with no recurrence signal.`;
 function buildScoringParagraph(has) {
   const base = "Two scoring layers: every lead has a basic `score` (firmographic \u2014 already decent, usually correlates with AI). Roughly the top 10 of each batch are also AI-qualified (targeted web research + qualification questions \u2192 `ai_agent_lead_score`, surfaced as `qualification_summary` on leadbay_pull_leads). Leads past the top ~10 are not worse \u2014 the system is saving resources.";
   const deepenTools = [];
@@ -22844,6 +22852,8 @@ function buildServerInstructions(exposed) {
   if (has("leadbay_agent_memory_capture")) {
     parts.push(AGENT_MEMORY);
   }
+  parts.push(ARTIFACT_PROPOSAL_PARAGRAPH);
+  parts.push(SCHEDULED_TASK_PARAGRAPH);
   parts.push(BUILTIN_WIDGETS_PARAGRAPH);
   return parts.join("\n\n");
 }
@@ -24638,7 +24648,7 @@ var OAUTH_BASE_URLS = {
     fr: "https://staging.api.leadbay.app"
   }
 };
-var VERSION = "0.18.1";
+var VERSION = "0.19.0";
 var HELP = `
 leadbay-mcp ${VERSION} \u2014 Leadbay Model Context Protocol server