kimaki 0.28.0 → 0.30.0

package/src/sleep.e2e.test.ts

@@ -344,19 +344,19 @@ describe('kimaki_sleep', () => {
         "--- from: user (sleep-tester)
         SLEEP_TOOL_MARKER wait for the deploy
         --- from: assistant (TestBot)
-        > *using deterministic-provider/deterministic-v2*
+        -# *using deterministic-provider/deterministic-v2*
         > waiting for the deploy
 
         ┣ kimaki_sleep until 2030-01-01T09:00:00Z _waiting for the deploy_
 
         > sleep-started
-        > *project ⋅ main ⋅ <1s ⋅ 0% ⋅ deterministic-v2*
-        ⬦ Woke after sleeping until 2030-01-01 09:00 UTC
+        -# *project ⋅ main ⋅ Ns ⋅ N% ⋅ deterministic-v2*
+        -# Woke after sleeping until 2030-01-01 09:00 UTC
         Reason: waiting for the deploy
         Continue the work you were waiting for.
         [embed]
         > sleep-wake-done
-        > *project ⋅ main ⋅ <1s ⋅ 0% ⋅ deterministic-v2* <@200000000000000991>"
+        -# *project ⋅ main ⋅ Ns ⋅ N% ⋅ deterministic-v2*"
       `)
 
       // The wake must continue the SAME OpenCode session, not start a new one:
@@ -465,16 +465,16 @@ describe('kimaki_sleep', () => {
         "--- from: user (sleep-tester)
         SLEEP_CANCEL_MARKER wait for something
         --- from: assistant (TestBot)
-        > *using deterministic-provider/deterministic-v2*
+        -# *using deterministic-provider/deterministic-v2*
         ┣ kimaki_sleep for 2h _cancel me_
 
         > cancel-sleep-started
-        > *project ⋅ main ⋅ <1s ⋅ 0% ⋅ deterministic-v2*
+        -# *project ⋅ main ⋅ Ns ⋅ N% ⋅ deterministic-v2*
         --- from: user (sleep-tester)
         SLEEP_CANCEL_FOLLOWUP never mind, keep going
         --- from: assistant (TestBot)
         > cancel-followup-done
-        > *project ⋅ main ⋅ <1s ⋅ 0% ⋅ deterministic-v2* <@200000000000000991>"
+        -# *project ⋅ main ⋅ Ns ⋅ N% ⋅ deterministic-v2*"
       `)
     },
     20_000,

package/src/store.ts

@@ -69,7 +69,7 @@ export type KimakiState = {
   critiqueEnabled: boolean
 
   // Whether final session footers mention the thread creator.
-  // Changes: set once at startup from --skip-footer-mentions.
+  // Changes: set once at startup from --enable-footer-mentions (default off).
   // Read by: ThreadSessionRuntime.emitFooter().
   footerMentionsEnabled: boolean
 
@@ -204,7 +204,7 @@ export const store = createStore<KimakiState>(() => ({
   defaultVerbosity: 'text_and_essential_tools',
   defaultMentionMode: false,
   critiqueEnabled: true,
-  footerMentionsEnabled: true,
+  footerMentionsEnabled: false,
   enabledSkills: [],
   disabledSkills: [],
   allowedMentions: ['users'],

package/src/system-message.test.ts

@@ -143,6 +143,17 @@ describe('system-message', () => {
     ).rejects.toMatchObject({ code: 'EISDIR' })
   })
 
+  test('tells agents to read compressed session transcripts under 100 KB', () => {
+    const message = getOpencodeSystemMessage({
+      sessionId: 'ses_123',
+      channelId: 'chan_123',
+      threadId: 'thread_123',
+    })
+    expect(message).toContain('kimaki session read <sessionId> > ./tmp/session.md 2>/dev/null')
+    expect(message).toContain('If it is under 100 KB, read the whole file. Do not grep first.')
+    expect(message).not.toContain('rg ')
+  })
+
   test('includes all-projects session search example', () => {
     const message = getOpencodeSystemMessage({
       sessionId: 'ses_123',
@@ -165,7 +176,7 @@ describe('system-message', () => {
       "kimaki session title 'Short title' --session ses_123",
     )
     expect(message).toContain(
-      'When the session scope or goal changed from the first message, update the title:',
+      'Skip the first turn. OpenCode already auto-generates the title from the first message.',
     )
     expect(message).toContain('thread-name="..."')
     expect(message).toContain('Do not retitle every turn')
@@ -415,14 +426,14 @@ describe('system-message', () => {
 
       ## updating the session title
 
-      When the session scope or goal changed from the first message, update the title:
+      Skip the first turn. OpenCode already auto-generates the title from the first message.
+      On later turns, if the scope or goal changed, run:
 
       kimaki session title 'Short title' --session ses_123
 
-      The current Discord thread title is in the per-turn \`<discord-user thread-name="..." />\` metadata.
-      This updates the OpenCode title. Discord follows automatically.
+      Current Discord title is in \`<discord-user thread-name="..." />\`. Discord follows the OpenCode title.
       Do not retitle every turn. Discord rate-limits thread renames.
-      Keep titles short. Do not add emoji. Do not copy ⬦, btw:, or Fork: prefixes.
+      Keep titles short. No emoji. No ⬦, btw:, or Fork: prefixes.
 
       ## discord user mentions
 
@@ -456,7 +467,17 @@ describe('system-message', () => {
 
       kimaki send --thread <thread_id> --prompt 'follow-up prompt' --agent <current_agent>
 
-      Use this when you already have the Discord thread ID. Prefer \`--thread\` over \`--session\` because thread IDs work across machines while session IDs only resolve on the machine that created the session.
+      Use this when you already have the Discord thread ID. Prefer \`--thread\` over \`--session\`. Discord thread IDs work on every computer. Session IDs do not.
+
+      A session ID only exists in the local SQLite database and OpenCode server on the computer that created it. \`kimaki send --session\` and \`kimaki session read\` fail on another computer.
+
+      To continue a remote thread, find the Discord channel, list its threads, then send with \`--thread\`:
+
+      \`\`\`bash
+      kimaki project list --all --json
+      kimaki thread list --channel <channel_id> --json
+      kimaki send --thread <thread_id> --prompt 'continue the work' --agent <current_agent>
+      \`\`\`
 
       To send to the thread associated with a known session (same machine only):
 
@@ -530,7 +551,7 @@ describe('system-message', () => {
 
       Scheduled tasks do not overlap by default. Add \`--allow-concurrency\` only when concurrent sessions from the same task are safe.
 
-      Pass \`--user\` when the user should see or act on the thread. Omit \`--user\` for routine autonomous work (digests, monitors, housekeeping) so the thread does not appear in the user's Discord sidebar. Discord only shows a thread in the left sidebar to its members. Without \`--user\`, kimaki does not ensure anyone is a member. This applies to \`--channel\` and \`--thread\` scheduling alike.
+      For autonomous scheduled tasks, omit \`--user\` by default so each run stays out of the user's Discord sidebar. Put the user's Discord ID in the task instructions instead, and tell the agent to mention that ID only when it completed work worth reviewing, found an issue, or needs a decision. Do not mention the user when the run found no work or made no changes. Reserve \`--user\` for reminders that must surface every time they fire.
 
       ALL scheduling is in UTC. Dates must be UTC ISO format ending with \`Z\`. Cron expressions also fire in UTC (e.g. \`0 9 * * 1\` means 9:00 UTC every Monday).
       When the user specifies a time without a timezone, ask them to confirm their timezone or the UTC equivalent. Never guess the user's timezone.
@@ -542,7 +563,7 @@ describe('system-message', () => {
       - \`--pre-run\` to start only when a project command exits with code 0
       - \`--allow-concurrency\` to permit overlapping runs from the same task
       - \`--parent-session\` to pass this session as parent of the scheduled child
-      - \`--user\` to add a specific user to the scheduled thread. Omit this for quiet autonomous work
+      - \`--user\` to add a specific user to every scheduled thread. Use this for reminders, not autonomous tasks
 
       \`--wait\` is incompatible with \`--send-at\` because scheduled tasks run in the future.
 
@@ -567,9 +588,9 @@ describe('system-message', () => {
 
       Notification strategy:
       - NEVER use \`@username\` (e.g. \`@Tommy\`) directly in task prompts. The prompt text becomes the first message in the thread, so a raw \`@\` mention triggers an actual Discord ping every time the task fires. Instead, wrap it in inline code like \`\\\`@Tommy\\\`\`, or use Discord user ID mentions like \`<@USER_ID>\` only in the body of the prompt where the agent will process it, not in the opening line.
-      - If a task needs user attention, add "mention the user via Discord user ID when task requires user review" in the task md file.
-      - With \`--user\`, the user is added to the thread and receives thread-level notifications. Omit \`--user\` when the work should stay out of the sidebar.
-      - If a scheduled task completes with no actionable result, archive the session: \`kimaki session archive thread_123 (or --session ses_123)\`
+      - For autonomous tasks, include the user's Discord ID in the task md file and instruct the agent to mention it only after completing work the user should review, when reporting an issue, or when asking for a decision.
+      - Do not mention the user if the task found no work, made no changes, or has nothing actionable to report. Archive that session instead: \`kimaki session archive thread_123 (or --session ses_123)\`
+      - Do not pass \`--user\` for autonomous tasks. It adds the user to every task thread before the result is known. Use it only when every occurrence must appear in the user's sidebar, such as an explicit reminder.
 
       Manage scheduled tasks with:
 
@@ -577,7 +598,7 @@ describe('system-message', () => {
       kimaki task edit <id> --prompt "new prompt" [--send-at "new schedule"] [--pre-run "command"] [--allow-concurrency true|false] [--user "<discord-user-id>"] [--model "provider/model"] [--agent "<agent>"]
       kimaki task delete <id>
 
-      \`kimaki task list\` prints \`userId\`, \`agent\`, and \`model\` columns. A \`-\` in \`userId\` means nobody is added to the thread when that task fires, so the user may never see it. Add a user with \`kimaki task edit <id> --user '<discord-user-id>'\`. Clear a stored user with \`kimaki task edit <id> --user ''\`. Change model or agent in place with \`--model\` / \`--agent\` (empty string clears the override). Do not read SQLite or recreate the task just to swap model.
+      \`kimaki task list\` prints \`userId\`, \`agent\`, and \`model\` columns. For autonomous tasks, \`userId\` should normally be \`-\`; put the Discord ID and conditional mention rule in the task instructions instead. Clear a stored user with \`kimaki task edit <id> --user ''\`. Add \`--user '<discord-user-id>'\` only for tasks whose every occurrence must surface. Change model or agent in place with \`--model\` / \`--agent\` (empty string clears the override). Do not read SQLite or recreate the task just to swap model.
 
       \`kimaki session list\` also shows if a session was started by a scheduled \`delay\` or \`cron\` task, including task ID when available.
 
@@ -656,10 +677,15 @@ describe('system-message', () => {
       kimaki session list --json  # machine-readable output
       kimaki session list --project /path/to/project  # specific project
 
+      # List sessions across every locally registered project
+      kimaki session list --all
+
       # List only in-progress sessions. Exit status is 1 when none remain.
       kimaki session list --active
       \`\`\`
 
+      Each row shows the session title, project directory, \`status: working\` or \`status: idle\`, and \`tokens: N\` (total token footprint) when available. Kimaki-started sessions also show their Discord \`thread\` ID.
+
       Titles prefixed with \`btw:\` are side sessions that answer a related user question in parallel. They are not duplicate sessions of the main task.
 
       To search past sessions (supports plain text or /regex/flags). Defaults to this project and the last 14 days. Use \`--days 0\` for all time. Use \`--all\` to search every locally registered project:
@@ -673,14 +699,13 @@ describe('system-message', () => {
       kimaki session search "auth timeout" --all
       \`\`\`
 
-      To read a session's full conversation as markdown, pipe to a file and grep it to avoid wasting context.
-      Logs go to stderr, so redirect stderr to hide them:
+      To read a session as markdown, pipe to a file. Logs go to stderr:
 
       \`\`\`bash
       kimaki session read <sessionId> > ./tmp/session.md 2>/dev/null
       \`\`\`
 
-      Then use grep/read tools on the file to find what you need.
+      The dump is already compressed (no thinking, truncated tool inputs). If it is under 100 KB, read the whole file. Do not grep first. Use \`--thinking\` / \`--verbose\` only when you need the full dump.
 
       ### who edited a file
 
@@ -709,18 +734,25 @@ describe('system-message', () => {
 
       When the user references another project by name, run \`kimaki project list\` to find its directory path and channel ID. Then read files, search code, or run commands directly in that directory. If the project is not listed, use \`kimaki project add /path/to/repo\` to register it and create a Discord channel for it. Do not add subfolders of an existing project — only add root project directories.
 
-      When the user uses \`#project-name\` syntax, they usually mean a Kimaki project channel. Use \`kimaki project list --json\` to resolve the \`channel_name\` to its repo working directory. The JSON output includes \`guild_id\` and \`guild_name\` to distinguish channels with the same name across different servers. When duplicates exist, prefer filtering by \`guild_id\` (stable) over \`guild_name\` (mutable): \`kimaki project list --json | jq -r '.[] | select(.channel_name == "project-name" and .guild_id == "123456") | .channel_id'\`.
+      When the user uses \`#project-name\` syntax, they usually mean a Kimaki project channel. Use \`kimaki project list --json\` to resolve the \`channel_name\` to its repo working directory. The JSON output includes \`guild_id\` and \`guild_name\` so you can tell channels with the same name apart. Prefer \`guild_id\` when there are duplicates. It is stable. \`guild_name\` can change.
 
-      When the user uses \`#Some Thread Title\` with spaces, they mean a **thread title**, not a project channel. Find the session by searching across projects, then read the session markdown:
+      If the local list has no match, the channel may live on another computer. Scan Discord with \`kimaki project list --all --json\` and send by \`channel_id\`. Remote rows have \`is_local: false\` and \`directory: null\`. Do not use \`--project\` or \`--session\` for those channels.
 
-      \`\`\`bash
-      # 1. Find the session ID by searching thread titles across all projects
-      kimaki session list --project /path/to/project --json | jq -r '.[] | select(.title | test("Thread Title"; "i")) | .id + " | " + .title'
+      When the user uses \`#Some Thread Title\` with spaces, they mean a **thread title**, not a project channel. On this computer, search local sessions, then read the markdown:
 
-      # 2. Read the full session conversation as markdown
+      \`\`\`bash
+      kimaki session list --project /path/to/project --json
       kimaki session read <sessionId> > ./tmp/session.md 2>/dev/null
       \`\`\`
 
+      If that local search has no match, list Discord threads instead of using a session ID:
+
+      \`\`\`bash
+      kimaki project list --all --json
+      kimaki thread list --channel <channel_id> --json
+      kimaki send --thread <thread_id> --prompt 'continue the work' --agent <current_agent>
+      \`\`\`
+
       If you don't know which project the thread belongs to, try each project from \`kimaki project list --json\`.
 
       \`\`\`bash
@@ -728,10 +760,18 @@ describe('system-message', () => {
       kimaki project list
       kimaki project list --json  # machine-readable output with guild_id, guild_name, is_local
 
-      # Include projects from other machines (scans Kimaki category in Discord)
+      # Include projects from other machines (scans Kimaki groups in Discord)
       kimaki project list --all
       kimaki project list --all --json  # remote projects have is_local: false and directory: null
 
+      # Find a channel on another computer, then send a new prompt there
+      kimaki project list --all --json
+      kimaki send --channel <channel_id> --prompt 'Plan how to update the API client to v2' --agent <current_agent>
+
+      # Continue an existing remote thread. Session IDs do not work across computers.
+      kimaki thread list --channel <channel_id> --json
+      kimaki send --thread <thread_id> --prompt 'continue the work' --agent <current_agent>
+
       # Resolve by channel name (prefer adding guild_name filter if duplicates exist)
       kimaki project list --json | jq -r '.[] | select(.channel_name == "project-name") | .channel_id + " " + .guild_name + " " + .directory'
 
@@ -793,7 +833,7 @@ describe('system-message', () => {
 
       \`session list --active\` exits with status 0 while it finds active sessions and status 1 when none remain. Exclude the current session in a wait loop so the loop does not wait for itself.
 
-      \`session wait\` exits with the session markdown on stdout once the model finishes responding.
+      \`session wait\` exits with the session markdown on stdout once the model finishes responding, or when the session pauses to show the user a question (it will not finish on its own until answered).
 
       Use \`--wait\` when you need to:
       - **Fix a bug in another project** before continuing here (e.g. fix a dependency, then resume)
@@ -1084,6 +1124,8 @@ describe('system-message', () => {
   test('moves per-turn discord metadata into synthetic prompt context', () => {
     expect(
       getOpencodePromptContext({
+        sessionId: 'ses_123',
+        threadId: 'thread_123',
         username: 'Tommy',
         userId: 'user_123',
         sourceMessageId: 'msg_123',
@@ -1104,6 +1146,11 @@ describe('system-message', () => {
     ).toMatchInlineSnapshot(`
       "<discord-user name="Tommy" user-id="user_123" message-id="msg_123" thread-id="thread_123" thread-name="Fix auth bug" />
 
+      <system-reminder>
+      Your current OpenCode session ID is: ses_123
+      Your current Discord thread ID is: thread_123
+      </system-reminder>
+
       This message was a reply to message
 
       <replied-message author="alice">

package/src/system-message.ts

@@ -392,6 +392,8 @@ function escapePromptText(value: string): string {
 }
 
 export function getOpencodePromptContext({
+  sessionId,
+  threadId,
   username,
   userId,
   sourceMessageId,
@@ -402,6 +404,8 @@ export function getOpencodePromptContext({
   currentAgent,
   worktreeChanged,
 }: {
+  sessionId?: string
+  threadId?: string
   username?: string
   userId?: string
   sourceMessageId?: string
@@ -429,6 +433,17 @@ export function getOpencodePromptContext({
       ? [` thread-name="${escapePromptAttribute(threadName)}"`]
       : []),
   ].join('')
+  const identityLines = [
+    ...(sessionId
+      ? [`Your current OpenCode session ID is: ${sessionId}`]
+      : []),
+    ...(threadId
+      ? [`Your current Discord thread ID is: ${threadId}`]
+      : []),
+  ]
+  const identityReminder = identityLines.length > 0
+    ? `<system-reminder>\n${identityLines.join('\n')}\n</system-reminder>`
+    : undefined
   const repliedMessageXml = repliedMessage
     ? `This message was a reply to message
 
@@ -438,6 +453,7 @@ ${escapePromptText(repliedMessage.text)}
     : undefined
   const sections = [
     ...(userAttrs ? [`<discord-user${userAttrs} />`] : []),
+    ...(identityReminder ? [identityReminder] : []),
     ...(repliedMessageXml ? [repliedMessageXml] : []),
     ...(currentAgent
       ? [`<system-reminder>\nCurrent agent: ${currentAgent}\n</system-reminder>`]
@@ -625,14 +641,14 @@ Different from \`kimaki session archive\` which hides the thread.
 
 ## updating the session title
 
-When the session scope or goal changed from the first message, update the title:
+Skip the first turn. OpenCode already auto-generates the title from the first message.
+On later turns, if the scope or goal changed, run:
 
 kimaki session title 'Short title' --session ${sessionId}
 
-The current Discord thread title is in the per-turn \`<discord-user thread-name="..." />\` metadata.
-This updates the OpenCode title. Discord follows automatically.
+Current Discord title is in \`<discord-user thread-name="..." />\`. Discord follows the OpenCode title.
 Do not retitle every turn. Discord rate-limits thread renames.
-Keep titles short. Do not add emoji. Do not copy ⬦, btw:, or Fork: prefixes.
+Keep titles short. No emoji. No ⬦, btw:, or Fork: prefixes.
 
 ## discord user mentions
 
@@ -668,7 +684,17 @@ To send a prompt to an existing thread instead of creating a new one:
 
 kimaki send --thread <thread_id> --prompt 'follow-up prompt' --agent <current_agent>
 
-Use this when you already have the Discord thread ID. Prefer \`--thread\` over \`--session\` because thread IDs work across machines while session IDs only resolve on the machine that created the session.
+Use this when you already have the Discord thread ID. Prefer \`--thread\` over \`--session\`. Discord thread IDs work on every computer. Session IDs do not.
+
+A session ID only exists in the local SQLite database and OpenCode server on the computer that created it. \`kimaki send --session\` and \`kimaki session read\` fail on another computer.
+
+To continue a remote thread, find the Discord channel, list its threads, then send with \`--thread\`:
+
+\`\`\`bash
+kimaki project list --all --json
+kimaki thread list --channel <channel_id> --json
+kimaki send --thread <thread_id> --prompt 'continue the work' --agent <current_agent>
+\`\`\`
 
 To send to the thread associated with a known session (same machine only):
 
@@ -738,7 +764,7 @@ Use \`--pre-run '<command>'\` to check whether a scheduled task should start. Ki
 
 Scheduled tasks do not overlap by default. Add \`--allow-concurrency\` only when concurrent sessions from the same task are safe.
 
-Pass \`--user\` when the user should see or act on the thread. Omit \`--user\` for routine autonomous work (digests, monitors, housekeeping) so the thread does not appear in the user's Discord sidebar. Discord only shows a thread in the left sidebar to its members. Without \`--user\`, kimaki does not ensure anyone is a member. This applies to \`--channel\` and \`--thread\` scheduling alike.
+For autonomous scheduled tasks, omit \`--user\` by default so each run stays out of the user's Discord sidebar. Put the user's Discord ID in the task instructions instead, and tell the agent to mention that ID only when it completed work worth reviewing, found an issue, or needs a decision. Do not mention the user when the run found no work or made no changes. Reserve \`--user\` for reminders that must surface every time they fire.
 
 ALL scheduling is in UTC. Dates must be UTC ISO format ending with \`Z\`. Cron expressions also fire in UTC (e.g. \`0 9 * * 1\` means 9:00 UTC every Monday).
 When the user specifies a time without a timezone, ask them to confirm their timezone or the UTC equivalent. Never guess the user's timezone.
@@ -750,7 +776,7 @@ When the user specifies a time without a timezone, ask them to confirm their tim
 - \`--pre-run\` to start only when a project command exits with code 0
 - \`--allow-concurrency\` to permit overlapping runs from the same task
 - \`--parent-session\` to pass this session as parent of the scheduled child
-- \`--user\` to add a specific user to the scheduled thread. Omit this for quiet autonomous work
+- \`--user\` to add a specific user to every scheduled thread. Use this for reminders, not autonomous tasks
 
 \`--wait\` is incompatible with \`--send-at\` because scheduled tasks run in the future.
 
@@ -775,9 +801,9 @@ For simple reminders and notifications (\`--notify-only\`), inline the prompt di
 
 Notification strategy:
 - NEVER use \`@username\` (e.g. \`@Tommy\`) directly in task prompts. The prompt text becomes the first message in the thread, so a raw \`@\` mention triggers an actual Discord ping every time the task fires. Instead, wrap it in inline code like \`\\\`@Tommy\\\`\`, or use Discord user ID mentions like \`<@USER_ID>\` only in the body of the prompt where the agent will process it, not in the opening line.
-- If a task needs user attention, add "mention the user via Discord user ID when task requires user review" in the task md file.
-- With \`--user\`, the user is added to the thread and receives thread-level notifications. Omit \`--user\` when the work should stay out of the sidebar.
-- If a scheduled task completes with no actionable result, archive the session: \`kimaki session archive ${archiveTarget}\`
+- For autonomous tasks, include the user's Discord ID in the task md file and instruct the agent to mention it only after completing work the user should review, when reporting an issue, or when asking for a decision.
+- Do not mention the user if the task found no work, made no changes, or has nothing actionable to report. Archive that session instead: \`kimaki session archive ${archiveTarget}\`
+- Do not pass \`--user\` for autonomous tasks. It adds the user to every task thread before the result is known. Use it only when every occurrence must appear in the user's sidebar, such as an explicit reminder.
 
 Manage scheduled tasks with:
 
@@ -785,7 +811,7 @@ kimaki task list
 kimaki task edit <id> --prompt "new prompt" [--send-at "new schedule"] [--pre-run "command"] [--allow-concurrency true|false] [--user "<discord-user-id>"] [--model "provider/model"] [--agent "<agent>"]
 kimaki task delete <id>
 
-\`kimaki task list\` prints \`userId\`, \`agent\`, and \`model\` columns. A \`-\` in \`userId\` means nobody is added to the thread when that task fires, so the user may never see it. Add a user with \`kimaki task edit <id> --user '<discord-user-id>'\`. Clear a stored user with \`kimaki task edit <id> --user ''\`. Change model or agent in place with \`--model\` / \`--agent\` (empty string clears the override). Do not read SQLite or recreate the task just to swap model.
+\`kimaki task list\` prints \`userId\`, \`agent\`, and \`model\` columns. For autonomous tasks, \`userId\` should normally be \`-\`; put the Discord ID and conditional mention rule in the task instructions instead. Clear a stored user with \`kimaki task edit <id> --user ''\`. Add \`--user '<discord-user-id>'\` only for tasks whose every occurrence must surface. Change model or agent in place with \`--model\` / \`--agent\` (empty string clears the override). Do not read SQLite or recreate the task just to swap model.
 
 \`kimaki session list\` also shows if a session was started by a scheduled \`delay\` or \`cron\` task, including task ID when available.
 
@@ -864,10 +890,15 @@ kimaki session list
 kimaki session list --json  # machine-readable output
 kimaki session list --project /path/to/project  # specific project
 
+# List sessions across every locally registered project
+kimaki session list --all
+
 # List only in-progress sessions. Exit status is 1 when none remain.
 kimaki session list --active
 \`\`\`
 
+Each row shows the session title, project directory, \`status: working\` or \`status: idle\`, and \`tokens: N\` (total token footprint) when available. Kimaki-started sessions also show their Discord \`thread\` ID.
+
 Titles prefixed with \`btw:\` are side sessions that answer a related user question in parallel. They are not duplicate sessions of the main task.
 
 To search past sessions (supports plain text or /regex/flags). Defaults to this project and the last ${SESSION_SEARCH_DEFAULT_DAYS} days. Use \`--days 0\` for all time. Use \`--all\` to search every locally registered project:
@@ -881,14 +912,13 @@ kimaki session search "/panic|crash/i" --channel <channel_id>
 kimaki session search "auth timeout" --all
 \`\`\`
 
-To read a session's full conversation as markdown, pipe to a file and grep it to avoid wasting context.
-Logs go to stderr, so redirect stderr to hide them:
+To read a session as markdown, pipe to a file. Logs go to stderr:
 
 \`\`\`bash
 kimaki session read <sessionId> > ./tmp/session.md 2>/dev/null
 \`\`\`
 
-Then use grep/read tools on the file to find what you need.
+The dump is already compressed (no thinking, truncated tool inputs). If it is under 100 KB, read the whole file. Do not grep first. Use \`--thinking\` / \`--verbose\` only when you need the full dump.
 
 ### who edited a file
 
@@ -917,18 +947,25 @@ If this session edited the file, use this session ID instead. If several files c
 
 When the user references another project by name, run \`kimaki project list\` to find its directory path and channel ID. Then read files, search code, or run commands directly in that directory. If the project is not listed, use \`kimaki project add /path/to/repo\` to register it and create a Discord channel for it. Do not add subfolders of an existing project — only add root project directories.
 
-When the user uses \`#project-name\` syntax, they usually mean a Kimaki project channel. Use \`kimaki project list --json\` to resolve the \`channel_name\` to its repo working directory. The JSON output includes \`guild_id\` and \`guild_name\` to distinguish channels with the same name across different servers. When duplicates exist, prefer filtering by \`guild_id\` (stable) over \`guild_name\` (mutable): \`kimaki project list --json | jq -r '.[] | select(.channel_name == "project-name" and .guild_id == "123456") | .channel_id'\`.
+When the user uses \`#project-name\` syntax, they usually mean a Kimaki project channel. Use \`kimaki project list --json\` to resolve the \`channel_name\` to its repo working directory. The JSON output includes \`guild_id\` and \`guild_name\` so you can tell channels with the same name apart. Prefer \`guild_id\` when there are duplicates. It is stable. \`guild_name\` can change.
 
-When the user uses \`#Some Thread Title\` with spaces, they mean a **thread title**, not a project channel. Find the session by searching across projects, then read the session markdown:
+If the local list has no match, the channel may live on another computer. Scan Discord with \`kimaki project list --all --json\` and send by \`channel_id\`. Remote rows have \`is_local: false\` and \`directory: null\`. Do not use \`--project\` or \`--session\` for those channels.
 
-\`\`\`bash
-# 1. Find the session ID by searching thread titles across all projects
-kimaki session list --project /path/to/project --json | jq -r '.[] | select(.title | test("Thread Title"; "i")) | .id + " | " + .title'
+When the user uses \`#Some Thread Title\` with spaces, they mean a **thread title**, not a project channel. On this computer, search local sessions, then read the markdown:
 
-# 2. Read the full session conversation as markdown
+\`\`\`bash
+kimaki session list --project /path/to/project --json
 kimaki session read <sessionId> > ./tmp/session.md 2>/dev/null
 \`\`\`
 
+If that local search has no match, list Discord threads instead of using a session ID:
+
+\`\`\`bash
+kimaki project list --all --json
+kimaki thread list --channel <channel_id> --json
+kimaki send --thread <thread_id> --prompt 'continue the work' --agent <current_agent>
+\`\`\`
+
 If you don't know which project the thread belongs to, try each project from \`kimaki project list --json\`.
 
 \`\`\`bash
@@ -936,10 +973,18 @@ If you don't know which project the thread belongs to, try each project from \`k
 kimaki project list
 kimaki project list --json  # machine-readable output with guild_id, guild_name, is_local
 
-# Include projects from other machines (scans Kimaki category in Discord)
+# Include projects from other machines (scans Kimaki groups in Discord)
 kimaki project list --all
 kimaki project list --all --json  # remote projects have is_local: false and directory: null
 
+# Find a channel on another computer, then send a new prompt there
+kimaki project list --all --json
+kimaki send --channel <channel_id> --prompt 'Plan how to update the API client to v2' --agent <current_agent>
+
+# Continue an existing remote thread. Session IDs do not work across computers.
+kimaki thread list --channel <channel_id> --json
+kimaki send --thread <thread_id> --prompt 'continue the work' --agent <current_agent>
+
 # Resolve by channel name (prefer adding guild_name filter if duplicates exist)
 kimaki project list --json | jq -r '.[] | select(.channel_name == "project-name") | .channel_id + " " + .guild_name + " " + .directory'
 
@@ -1001,7 +1046,7 @@ while kimaki session list --active --exclude ${sessionId}; do sleep 5; done
 
 \`session list --active\` exits with status 0 while it finds active sessions and status 1 when none remain. Exclude the current session in a wait loop so the loop does not wait for itself.
 
-\`session wait\` exits with the session markdown on stdout once the model finishes responding.
+\`session wait\` exits with the session markdown on stdout once the model finishes responding, or when the session pauses to show the user a question (it will not finish on its own until answered).
 
 Use \`--wait\` when you need to:
 - **Fix a bug in another project** before continuing here (e.g. fix a dependency, then resume)

package/src/task-id-plugin.ts

@@ -0,0 +1,17 @@
+import type { Plugin } from '@opencode-ai/plugin'
+
+// Work around https://github.com/anomalyco/opencode/issues/49599 until invalid task IDs are ignored upstream.
+export function sanitizeTaskArguments(args: Record<string, unknown>) {
+  const taskId = args.task_id
+  if (typeof taskId !== 'string' || taskId.startsWith('ses')) return
+  delete args.task_id
+}
+
+export const taskIdPlugin: Plugin = async () => {
+  return {
+    'tool.execute.before': async (input, output) => {
+      if (input.tool !== 'task') return
+      sanitizeTaskArguments(output.args)
+    },
+  }
+}

package/src/task-runner.ts

@@ -455,6 +455,9 @@ async function executeScheduledTask({
   task: ScheduledTask
   runId?: number
 }): Promise<string | null | Error | { kind: 'condition-not-met' }> {
+  const claimedTask = await getScheduledTask(task.id)
+  if (claimedTask?.status !== 'running') return { kind: 'condition-not-met' }
+
   const payloadResult = parseScheduledTaskPayload(task.payload_json)
   if (payloadResult instanceof Error) {
     return new Error(`Task ${task.id} has invalid payload`, {
@@ -466,6 +469,11 @@ async function executeScheduledTask({
   if (commandResult instanceof Error) return commandResult
   if (commandResult.kind === 'skip') return { kind: 'condition-not-met' }
 
+  const taskAfterCommand = await getScheduledTask(task.id)
+  if (taskAfterCommand?.status !== 'running') {
+    return { kind: 'condition-not-met' }
+  }
+
   if (payloadResult.kind === 'thread') {
     return executeThreadScheduledTask({
       rest,

package/src/task-schedule.test.ts

@@ -210,7 +210,7 @@ describe('formatSessionSleepToolOutput', () => {
         reason: 'Wait 1 day, then check mail',
       }),
     ).toBe(
-      '⬦ Woke after sleeping until 2026-09-07 13:21 UTC\nReason: Wait 1 day, then check mail\nContinue the work you were waiting for.',
+      '-# Woke after sleeping until 2026-09-07 13:21 UTC\nReason: Wait 1 day, then check mail\nContinue the work you were waiting for.',
     )
   })
 })

package/src/test-utils.ts

@@ -354,9 +354,11 @@ export function isFooterMessage({
   if (message.author.id !== botUserId) {
     return false
   }
-  const content = message.content.startsWith('> ')
-    ? message.content.slice(2)
-    : message.content
+  const content = message.content.startsWith('-# ')
+    ? message.content.slice(3)
+    : message.content.startsWith('> ')
+      ? message.content.slice(2)
+      : message.content
   if (!content.startsWith('*')) {
     return false
   }