@jeffober/thatch 0.1.25 → 0.1.27

package/src/index.ts

@@ -19,7 +19,7 @@ import { installSkills, SHARED_SKILLS, OPENCODE_ONLY_SKILLS } from "./skills";
 import { hygieneReport } from "./hygiene";
 
 // ---------------------------------------------------------------------------
-// V1 server export — tools, prompt injection, session hooks
+// V1 server export - tools, prompt injection, session hooks
 // ---------------------------------------------------------------------------
 
 // Minimum cosine score for the prompt-aware recall nudge. Lower than
@@ -27,7 +27,7 @@ import { hygieneReport } from "./hygiene";
 // than "duplicate." Tunable via THATCH_RECALL_THRESHOLD.
 const RECALL_THRESHOLD = parseFloat(process.env.THATCH_RECALL_THRESHOLD ?? "0.55");
 
-// Prompts shorter than this skip the recall nudge — trivially short prompts
+// Prompts shorter than this skip the recall nudge - trivially short prompts
 // like "yes" or "ok" match too broadly to be useful.
 const MIN_PROMPT_LEN = 10;
 
@@ -50,12 +50,12 @@ export const server: Plugin = async ({ client, worktree }) => {
   const extraction = new ExtractionPipeline();
 
   // Sessions currently being compacted. chat.message nudges are skipped while
-  // a session is in this set — the agent can't call tools during summary
+  // a session is in this set - the agent can't call tools during summary
   // generation, so a recall or extraction nudge would cause a blocked-tool
   // error. Cleared by experimental.compaction.autocontinue (success), the
   // session.compacted event (redundant belt-and-suspenders), or chat.message
   // itself when a non-compaction message arrives (compaction failure fallback
-  // — autocontinue never fired, so the next real user message clears the
+  // - autocontinue never fired, so the next real user message clears the
   // stale flag). If all three somehow miss, the flag leaks (graceful
   // degradation: nudges stay off for that session, but no crash).
   const compacting = new Set<string>();
@@ -78,7 +78,7 @@ export const server: Plugin = async ({ client, worktree }) => {
   // When the child writes a memory, consumeSnapshot drains only these entries
   // (by reference identity), preserving interleaved-turn entries that arrived
   // while the sub-agent was running. Without this, the child's memory_remember
-  // would drain the parent's ENTIRE buffer — silently dropping facts from
+  // would drain the parent's ENTIRE buffer - silently dropping facts from
   // any tool calls the parent made concurrently with the sub-agent.
   const parentSnapshots = new Map<string, ToolInteraction[]>();
 
@@ -108,7 +108,7 @@ export const server: Plugin = async ({ client, worktree }) => {
   // Keyed by child session ID. Cleaned up on child idle, error, or deletion.
   const childMetrics = new Map<string, { new: number; updated: number; deleted: number }>();
 
-  // Skills always install to the global opencode config — installing into the
+  // Skills always install to the global opencode config - installing into the
   // worktree would mutate the user's repo (untracked files in git status).
   // A failed install degrades the nudge workflow but must not kill the plugin.
   try {
@@ -129,7 +129,7 @@ export const server: Plugin = async ({ client, worktree }) => {
   // drain machinery handles buffer cleanup. The nudge path is a fallback if
   // this throws.
   //
-  // Does NOT call extraction.accept — entries stay in pending so consumeSnapshot
+  // Does NOT call extraction.accept - entries stay in pending so consumeSnapshot
   // can drain them by reference identity when the child writes a memory. The
   // extracting set (not accept) suppresses the nudge in chat.message.
   //
@@ -183,7 +183,7 @@ export const server: Plugin = async ({ client, worktree }) => {
         throw err;
       }
     } else {
-      // Fire and forget — the parent is already idle, so blocking the event
+      // Fire and forget - the parent is already idle, so blocking the event
       // handler would only delay other event processing. The child runs to
       // completion and its idle event triggers cleanup.
       client.session
@@ -203,12 +203,12 @@ export const server: Plugin = async ({ client, worktree }) => {
   return {
     tool: createTools(db, model, repo),
 
-    // 1. System prompt — always in context.
+    // 1. System prompt - always in context.
     "experimental.chat.system.transform": async (_input, output) => {
       output.system.push(sys);
     },
 
-    // 2. Compaction context — re-familiarizes after compaction. The flag
+    // 2. Compaction context - re-familiarizes after compaction. The flag
     //    suppresses chat.message nudges during summary generation (tool
     //    calls are blocked there).
     "experimental.session.compacting": async (input, output) => {
@@ -222,15 +222,15 @@ export const server: Plugin = async ({ client, worktree }) => {
       compacting.delete(input.sessionID);
     },
 
-    // 3. Tool buffering — feeds the extraction pipeline (direct extraction
+    // 3. Tool buffering - feeds the extraction pipeline (direct extraction
     //    via SDK or nudge fallback). Excluded tools:
     //    - thatch_*: extracting facts from memory ops would echo the store
     //    - skill/task: meta-tools that orchestrate agent behavior (loading
     //      skills, dispatching sub-agents). Buffering them creates a feedback
-    //      loop — extraction triggers a skill load, which gets buffered, which
+    //      loop - extraction triggers a skill load, which gets buffered, which
     //      triggers another extraction on the next turn.
     //    Memory writes consume the buffer and reset the missed-nudge counter.
-    //    The buffer is NOT drained on nudge delivery — it persists until the
+    //    The buffer is NOT drained on nudge delivery - it persists until the
     //    agent writes a memory, so ignored nudges accumulate and the payload
     //    grows with each missed cycle.
     //
@@ -247,7 +247,7 @@ export const server: Plugin = async ({ client, worktree }) => {
     //      session errors or is deleted before completing.
     //
     //    A memory_remember call in a child session (sub-agent) also drains
-    //    the parent's pending buffer — but only the entries that existed at
+    //    the parent's pending buffer - but only the entries that existed at
     //    dispatch time (the snapshot). Entries from interleaved turns survive
     //    so their facts aren't silently dropped.
     "tool.execute.after": async (input, output) => {
@@ -269,7 +269,7 @@ export const server: Plugin = async ({ client, worktree }) => {
           // from the pending buffer. If no snapshot was recorded (unreachable
           // when childToParent has the entry, since both are set together in
           // session.created), skip the drain rather than dropping the entire
-          // buffer — interleaved-turn entries that arrived while the child
+          // buffer - interleaved-turn entries that arrived while the child
           // was running must survive for the next extraction cycle.
           extraction.completeAccepted(parentID);
           const snapshot = parentSnapshots.get(input.sessionID);
@@ -300,7 +300,7 @@ export const server: Plugin = async ({ client, worktree }) => {
         return;
       }
       // Track memory deletions in child sessions for the toast metrics.
-      // Same scoping as the remember handler above — only child sessions,
+      // Same scoping as the remember handler above - only child sessions,
       // not the parent or manual memory writes from the user's session.
       if (input.tool === "thatch_memory_forget" && childToParent.has(input.sessionID)) {
         const metrics = childMetrics.get(input.sessionID) ?? { new: 0, updated: 0, deleted: 0 };
@@ -318,7 +318,7 @@ export const server: Plugin = async ({ client, worktree }) => {
       });
     },
 
-    // 4. Per-message nudge — two priority tiers:
+    // 4. Per-message nudge - two priority tiers:
     //   a. Extraction nudge: prior tool interactions are queued for fact
     //      extraction (carries the JSON payload for thatch-fact-extractor).
     //   b. Recall + prediction nudge: when no extraction is pending, embed
@@ -337,17 +337,17 @@ export const server: Plugin = async ({ client, worktree }) => {
       if (compacting.has(input.sessionID)) {
         // The session is marked as compacting. If this message is the
         // compaction summary generation itself (has a compaction-type part),
-        // suppress nudges — tools are blocked during summary generation and a
+        // suppress nudges - tools are blocked during summary generation and a
         // nudge would cause a blocked-tool error. If it is NOT a compaction
         // message, compaction has already failed (autocontinue never fired)
         // and the user is sending a new message. Clear the stale flag and
-        // proceed normally — tools are available again.
+        // proceed normally - tools are available again.
         const isCompactionMsg = (output.parts as any[]).some((p) => p.type === "compaction");
         if (isCompactionMsg) return;
         compacting.delete(input.sessionID);
       }
       // Extraction nudge (fallback path). Skipped when the extracting set
-      // is active — that means a direct-extraction child session is running
+      // is active - that means a direct-extraction child session is running
       // and the plugin is handling extraction via the SDK. The nudge fires
       // here only when direct extraction was never triggered or threw.
       if (!extracting.has(input.sessionID) && extraction.pending(input.sessionID)) {
@@ -368,7 +368,7 @@ export const server: Plugin = async ({ client, worktree }) => {
         return;
       }
 
-      // No extraction pending — try the prompt-aware recall nudge. Extract
+      // No extraction pending - try the prompt-aware recall nudge. Extract
       // the user's prompt text from the message parts, embed it with the
       // warm in-process model, and search for matches. Best-effort: any
       // failure (no text, model not loaded, empty store) silently skips.
@@ -469,7 +469,7 @@ export const server: Plugin = async ({ client, worktree }) => {
         if (info.parentID) {
           childToParent.set(info.id, info.parentID);
           parentSnapshots.set(info.id, [...extraction.peek(info.parentID)]);
-          // Child sessions don't need the session-start reminder — only
+          // Child sessions don't need the session-start reminder - only
           // top-level sessions do. Returning here prevents extraction
           // children from receiving the hygiene report and tool overview.
           return;
@@ -499,23 +499,23 @@ export const server: Plugin = async ({ client, worktree }) => {
           // - Task-dispatched sub-agent (code review specialist, nudge-path
           //   fact-extractor, any model-dispatched task): complete the
           //   parent's accepted entries and reset missedNudges. Do NOT drain
-          //   the buffer or delete the session — the task tool that
+          //   the buffer or delete the session - the task tool that
           //   dispatched the sub-agent needs to read its output.
           if (extractionChildren.has(sessionID)) {
             // Drain the parent's snapshot entries from the pending buffer.
             // If the child wrote memories, consumeSnapshot already ran in
-            // tool.execute.after and the snapshot is gone — nothing to
+            // tool.execute.after and the snapshot is gone - nothing to
             // drain. If the child did a no-save run, the snapshot entries
             // are still in the buffer and need to be drained here so they
             // don't replay as a nudge on the next chat.message. Never drain
-            // the entire buffer — interleaved-turn entries must survive.
+            // the entire buffer - interleaved-turn entries must survive.
             const snapshot = parentSnapshots.get(sessionID);
             if (snapshot) {
               extraction.consumeSnapshot(parentID, snapshot);
             }
 
             // Fire a toast with the extraction metrics. Only show a toast
-            // when memories were actually written — no toast for no-save
+            // when memories were actually written - no toast for no-save
             // runs to avoid notification fatigue.
             const metrics = childMetrics.get(sessionID);
             const parts: string[] = [];
@@ -546,7 +546,7 @@ export const server: Plugin = async ({ client, worktree }) => {
             try {
               await client.session.delete({ path: { id: sessionID } });
             } catch {
-              // Best-effort — the child is idle and harmless if not deleted.
+              // Best-effort - the child is idle and harmless if not deleted.
             }
           } else {
             // Task-dispatched sub-agent went idle. Complete the parent's
@@ -557,7 +557,7 @@ export const server: Plugin = async ({ client, worktree }) => {
           }
           return;
         }
-        // Parent went idle — trigger direct extraction if there are pending
+        // Parent went idle - trigger direct extraction if there are pending
         // tool interactions and no extraction is already running. Falls
         // back to the nudge path on the next chat.message if this throws.
         if (

package/src/mcp.ts

@@ -39,14 +39,14 @@ const SERVER_VERSION = pkg.version;
 
 /**
  * MCP capabilities declared in the initialize response. Thatch is a tools-only
- * server — no resources, prompts, or subscriptions.
+ * server - no resources, prompts, or subscriptions.
  */
 const CAPABILITIES = {
   tools: { listChanged: false },
 };
 
 // ---------------------------------------------------------------------------
-// Tool index — builds validators and JSON Schema once at server startup
+// Tool index - builds validators and JSON Schema once at server startup
 // ---------------------------------------------------------------------------
 
 interface CompiledTool {
@@ -74,7 +74,7 @@ function compileTools(): Map<string, CompiledTool> {
 
 /**
  * Runs the thatch MCP server over stdio. Reads newline-delimited JSON-RPC from
- * stdin, writes responses to stdout. All diagnostics go to stderr — any stray
+ * stdin, writes responses to stdout. All diagnostics go to stderr - any stray
  * output on stdout corrupts the protocol.
  *
  * Repo identity is resolved from CLAUDE_PROJECT_DIR (set by Claude Code for
@@ -113,7 +113,7 @@ export async function runMcpServer(): Promise<void> {
   }
 
   // The sideband socket lets one-shot hook processes (flush-tools) ask the
-  // warm MCP server to embed a prompt and search for matches — without
+  // warm MCP server to embed a prompt and search for matches - without
   // loading the ~34 MB model themselves. If the socket can't be opened
   // (permissions, tmpdir issues), the MCP server still works; only the
   // prompt-aware recall nudge degrades.
@@ -146,7 +146,7 @@ export async function runMcpServer(): Promise<void> {
         continue;
       }
 
-      // Notifications (no id) are fire-and-forget — no response expected.
+      // Notifications (no id) are fire-and-forget - no response expected.
       const isNotification = req.id === null || req.id === undefined;
       const res = await dispatch(req, tools, ctx);
       if (res === null || isNotification) continue;

package/src/prompts.ts

@@ -37,18 +37,20 @@ you're working in. Prior sessions may have already investigated it.
 Thatch ships skills for code review, project investigation, and memory
 workflows. The host auto-discovers them, but reach for them proactively:
 
-- \`thatch-code-review\` — full multi-agent code review (dispatches 7 specialists + synthesizer). Requires sub-agent support.
-- \`thatch-workflow-research\` — research code workflows affected by a change or planned change, before reviewing or planning.
-- \`thatch-change-walkthrough\` — produces a teaching walkthrough of a diff with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + change overlay mirroring the same numbers) and calibrated prose rules. Load the skill for the format and prose rules, not just for the research method.
-- \`thatch-code-walkthrough\` — produces a teaching walkthrough for a feature/workflow as it stands today with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + Key files) and calibrated prose rules. Also use proactively to draft high-level docs for new or undocumented features (suitable for pkg/README.md or docs/features/<name>.md). Load the skill for the format and prose rules, not just for the research method.
-- \`thatch-review-context\` — gather project context (PRs, tickets, TODOs, deferred work) before a review.
-- \`thatch-project-primer\` — investigate a new project and write foundational memories.
-- \`thatch-session-reflection\` — record what you learned at end of session.
-- \`thatch-fact-extractor\` — extract durable facts from recent tool interactions (auto-triggered by the extraction pipeline).
-- \`thatch-dedup-classifier\` — resolve duplicate memory pairs from \`thatch_find_duplicates\`.
-- \`pr-description\` — draft a PR description with SYNOPSIS / PURPOSE / DESCRIPTION / WALK-THROUGH / NOTES, using bold and italic emphasis on the phrases that carry the meaning.
-- \`ticket-description\` — draft a ticket, issue, or work item (Linear or Jira) with clear sections and bold/italic emphasis for scanning.
-- \`split-overlarge-pr\` — split already-completed work from an overlarge PR into human-reviewable, release-safe PRs targeting main.
+- \`thatch-code-review\` - full multi-agent code review (dispatches 7 specialists + synthesizer). Requires sub-agent support.
+- \`thatch-review-followup\` - alternate entrypoint for follow-up review rounds. Verifies whether the author's responses and code changes since your last review adequately addressed prior findings, offers to reply on resolved items, then optionally re-runs the full structured review.
+- \`thatch-code-archaeology\` - investigate an existing feature, debug an unfamiliar area, or begin a new ticket. Explores the code base from multiple angles (data model, state flow, git history, sibling features) before proposing changes. The research skill; pair with \`thatch-coding-workflow\`.
+- \`thatch-change-walkthrough\` - produces a teaching walkthrough of a diff with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + change overlay mirroring the same numbers) and calibrated prose rules. Load the skill for the format and prose rules, not just for the research method.
+- \`thatch-code-walkthrough\` - produces a teaching walkthrough for a feature/workflow as it stands today with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + Key files) and calibrated prose rules. Also use proactively to draft high-level docs for new or undocumented features (suitable for pkg/README.md or docs/features/<name>.md). Load the skill for the format and prose rules, not just for the research method.
+- \`thatch-review-context\` - gather project context (PRs, tickets, TODOs, deferred work) before a review.
+- \`thatch-project-primer\` - investigate a new project and write foundational memories.
+- \`thatch-session-reflection\` - record what you learned at end of session.
+- \`thatch-fact-extractor\` - extract durable facts from recent tool interactions (auto-triggered by the extraction pipeline).
+- \`thatch-dedup-classifier\` - resolve duplicate memory pairs from \`thatch_find_duplicates\`.
+- \`thatch-pr-description\` - draft a PR description with SYNOPSIS / PURPOSE / DESCRIPTION / WALK-THROUGH / NOTES, using bold and italic emphasis on the phrases that carry the meaning.
+- \`thatch-ticket-description\` - draft a ticket, issue, or work item (Linear or Jira) with clear sections and bold/italic emphasis for scanning.
+- \`thatch-split-overlarge-pr\` - split already-completed work from an overlarge PR into human-reviewable, release-safe PRs targeting main.
+- \`thatch-review-response\` - respond to code review on the user's own PR. Triage findings, fix bugs one by one, reply on each thread, then post a top-level summary comment.
 
 ## When to Write
 
@@ -60,8 +62,8 @@ current state of the code).
 
 ## Before Responding
 
-Before composing a final response after substantial work this turn — multiple
-rounds of tool calls for investigation, debugging, or code-writing — check
+Before composing a final response after substantial work this turn - multiple
+rounds of tool calls for investigation, debugging, or code-writing - check
 whether you've discovered knowledge worth persisting. Use thatch_memory_recall
 to check for duplicates, then thatch_memory_remember for new findings. Then
 deliver your response.
@@ -69,7 +71,7 @@ deliver your response.
 ## Background Task Completions
 
 When a background task completes and its result is injected into the session,
-the completion is informational — it is not user input and not approval to
+the completion is informational - it is not user input and not approval to
 proceed with anything you were discussing. Do not narrate what the task did or
 summarize its output. If you were mid-conversation with the user, continue
 that thread. If the user's last message has been fully addressed, stop and
@@ -78,7 +80,7 @@ not greenlit.
 
 If a background task was cancelled after timing out, re-dispatch it once with
 the same prompt. If the re-dispatch also fails or times out, note the gap and
-move on. Do not retry more than once — repeated failures indicate an
+move on. Do not retry more than once - repeated failures indicate an
 intractable prompt, not a transient hang.
 
 ## User Decision Model
@@ -146,7 +148,7 @@ do not save it. Do not store:
 
 ## Archived Memories
 
-Memories can be marked \`archived\` — a flag for stable, long-term historical
+Memories can be marked \`archived\` - a flag for stable, long-term historical
 records that should not trigger hygiene nudges (stale, orphaned, duplicate).
 Archived memories are excluded from search/recall results by default; pass
 \`includeArchived: true\` to surface them for archaeological dives.
@@ -154,25 +156,25 @@ Archived memories are excluded from search/recall results by default; pass
 When a branch is merged or about to be deleted, consolidate its branch-scoped
 memories into a single archived memory scoped to that same branch (preserving
 provenance). Capture intent, design decisions, review back-and-forth, PR
-number, unexpected pivots — the kind of git-archaeology context that explains
+number, unexpected pivots - the kind of git-archaeology context that explains
 _why_ the code looks the way it does a year from now. Then memory_forget the
 originals. The archived record outlives the branch. Future sessions searching
 with \`includeArchived: true\` can pull it up when investigating ambiguous code.
 
-Updating an archived memory requires passing \`archived\` explicitly — omit it
+Updating an archived memory requires passing \`archived\` explicitly - omit it
 and the tool returns an error. Pass \`archived: true\` to keep it archived,
 \`archived: false\` to unarchive.
 
 ## Explicit Requests
 
-"Remember X" — save immediately.
-"Forget X" — thatch_memory_recall to find it, then thatch_memory_forget.`;
+"Remember X" - save immediately.
+"Forget X" - thatch_memory_recall to find it, then thatch_memory_forget.`;
 }
 
 export function compactionContext(repo: string): string {
   return `Thatch persistent memory is active with stores "${repo}" and "global".
 When summarizing, preserve context about what's been learned and decided
-this session — conventions, architectural decisions, and prior knowledge
+this session - conventions, architectural decisions, and prior knowledge
 retrieved from memory. This ensures continuity after compaction.`;
 }
 
@@ -198,7 +200,7 @@ When convenient this session (not before addressing the user's request), tend th
  * start. The repo store name is auto-detected at runtime by the MCP server, so
  * the instructions don't need it.
  *
- * Tool names use the bare form (memory_remember, memory_recall, etc.) — in
+ * Tool names use the bare form (memory_remember, memory_recall, etc.) - in
  * Claude Code they're prefixed as mcp__thatch__memory_remember, but the bare
  * names are what the agent searches for and what the instructions reference.
  */
@@ -239,18 +241,20 @@ you're working in. Prior sessions may have already investigated it.
 Thatch ships skills for code review, project investigation, and memory
 workflows. The host auto-discovers them, but reach for them proactively:
 
-- \`thatch-workflow-research\` — research code workflows affected by a change or planned change, before reviewing or planning.
-- \`thatch-change-walkthrough\` — produces a teaching walkthrough of a diff with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + change overlay mirroring the same numbers) and calibrated prose rules. Load the skill for the format and prose rules, not just for the research method.
-- \`thatch-code-walkthrough\` — produces a teaching walkthrough for a feature/workflow as it stands today with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + Key files) and calibrated prose rules. Also use proactively to draft high-level docs for new or undocumented features (suitable for pkg/README.md or docs/features/<name>.md). Load the skill for the format and prose rules, not just for the research method.
-- \`thatch-review-context\` — gather project context (PRs, tickets, TODOs, deferred work) before a review.
-- \`thatch-review-pedantic\` / \`-acceptance\` / \`-state-flow\` / \`-no-slop\` / \`-breadcrumbs\` / \`-mark-and-sweep\` / \`-highlights\` — seven specialist review lenses. Run individually, then \`thatch-review-synthesizer\` to verify and aggregate.
-- \`thatch-project-primer\` — investigate a new project and write foundational memories.
-- \`thatch-session-reflection\` — record what you learned at end of session.
-- \`thatch-fact-extractor\` — extract durable facts from recent tool interactions (auto-triggered by the extraction pipeline).
-- \`thatch-dedup-classifier\` — resolve duplicate memory pairs from \`find_duplicates\`.
-- \`pr-description\` — draft a PR description with SYNOPSIS / PURPOSE / DESCRIPTION / WALK-THROUGH / NOTES, using bold and italic emphasis on the phrases that carry the meaning.
-- \`ticket-description\` — draft a ticket, issue, or work item (Linear or Jira) with clear sections and bold/italic emphasis for scanning.
-- \`split-overlarge-pr\` — split already-completed work from an overlarge PR into human-reviewable, release-safe PRs targeting main.
+- \`thatch-code-archaeology\` - investigate an existing feature, debug an unfamiliar area, or begin a new ticket. Explores the code base from multiple angles (data model, state flow, git history, sibling features) before proposing changes. The research skill; pair with \`thatch-coding-workflow\`.
+- \`thatch-change-walkthrough\` - produces a teaching walkthrough of a diff with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + change overlay mirroring the same numbers) and calibrated prose rules. Load the skill for the format and prose rules, not just for the research method.
+- \`thatch-code-walkthrough\` - produces a teaching walkthrough for a feature/workflow as it stands today with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + Key files) and calibrated prose rules. Also use proactively to draft high-level docs for new or undocumented features (suitable for pkg/README.md or docs/features/<name>.md). Load the skill for the format and prose rules, not just for the research method.
+- \`thatch-review-context\` - gather project context (PRs, tickets, TODOs, deferred work) before a review.
+- \`thatch-review-followup\` - alternate entrypoint for follow-up review rounds. Verifies whether the author's responses and code changes since your last review adequately addressed prior findings, offers to reply on resolved items, then optionally re-runs the full structured review.
+- \`thatch-review-pedantic\` / \`-acceptance\` / \`-state-flow\` / \`-no-slop\` / \`-breadcrumbs\` / \`-mark-and-sweep\` / \`-highlights\` - seven specialist review lenses. Run individually, then \`thatch-review-synthesizer\` to verify and aggregate.
+- \`thatch-project-primer\` - investigate a new project and write foundational memories.
+- \`thatch-session-reflection\` - record what you learned at end of session.
+- \`thatch-fact-extractor\` - extract durable facts from recent tool interactions (auto-triggered by the extraction pipeline).
+- \`thatch-dedup-classifier\` - resolve duplicate memory pairs from \`find_duplicates\`.
+- \`thatch-pr-description\` - draft a PR description with SYNOPSIS / PURPOSE / DESCRIPTION / WALK-THROUGH / NOTES, using bold and italic emphasis on the phrases that carry the meaning.
+- \`thatch-ticket-description\` - draft a ticket, issue, or work item (Linear or Jira) with clear sections and bold/italic emphasis for scanning.
+- \`thatch-split-overlarge-pr\` - split already-completed work from an overlarge PR into human-reviewable, release-safe PRs targeting main.
+- \`thatch-review-response\` - respond to code review on the user's own PR. Triage findings, fix bugs one by one, reply on each thread, then post a top-level summary comment.
 
 ## When to Write
 
@@ -262,8 +266,8 @@ current state of the code).
 
 ## Before Responding
 
-Before composing a final response after substantial work this turn — multiple
-rounds of tool calls for investigation, debugging, or code-writing — check
+Before composing a final response after substantial work this turn - multiple
+rounds of tool calls for investigation, debugging, or code-writing - check
 whether you've discovered knowledge worth persisting. Use memory_recall to
 check for duplicates, then memory_remember for new findings. Then deliver
 your response.
@@ -333,7 +337,7 @@ do not save it. Do not store:
 
 ## Archived Memories
 
-Memories can be marked \`archived\` — a flag for stable, long-term historical
+Memories can be marked \`archived\` - a flag for stable, long-term historical
 records that should not trigger hygiene nudges (stale, orphaned, duplicate).
 Archived memories are excluded from search/recall results by default; pass
 \`includeArchived: true\` to surface them for archaeological dives.
@@ -341,19 +345,19 @@ Archived memories are excluded from search/recall results by default; pass
 When a branch is merged or about to be deleted, consolidate its branch-scoped
 memories into a single archived memory scoped to that same branch (preserving
 provenance). Capture intent, design decisions, review back-and-forth, PR
-number, unexpected pivots — the kind of git-archaeology context that explains
+number, unexpected pivots - the kind of git-archaeology context that explains
 _why_ the code looks the way it does a year from now. Then memory_forget the
 originals. The archived record outlives the branch. Future sessions searching
 with \`includeArchived: true\` can pull it up when investigating ambiguous code.
 
-Updating an archived memory requires passing \`archived\` explicitly — omit it
+Updating an archived memory requires passing \`archived\` explicitly - omit it
 and the tool returns an error. Pass \`archived: true\` to keep it archived,
 \`archived: false\` to unarchive.
 
 ## Explicit Requests
 
-"Remember X" — save immediately.
-"Forget X" — \`memory_recall\` to find it, then \`memory_forget\`.`;
+"Remember X" - save immediately.
+"Forget X" - \`memory_recall\` to find it, then \`memory_forget\`.`;
 }
 
 /**
@@ -401,7 +405,7 @@ export function extractionNudge(
   const plural = count === 1 ? "" : "s";
   // opencode's task tool has a `background` parameter (requires
   // OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS=true) that launches the sub-agent
-  // asynchronously — the call returns immediately and the session is notified
+  // asynchronously - the call returns immediately and the session is notified
   // on completion. Naming the parameter gives the model a direct hit on the tool
   // schema it sees. MCP hosts (Claude Code, Cursor) keep generic sub-agent
   // wording because their background mechanisms differ and are not exposed as a
@@ -441,7 +445,7 @@ export function extractionNudge(
  * Prompt text sent directly to a child session the plugin created for
  * extraction. Unlike extractionNudge (which tells the model to dispatch a
  * sub-agent and call extraction_done), this prompt is already inside the
- * child — the model just runs the fact-extractor skill and saves facts.
+ * child - the model just runs the fact-extractor skill and saves facts.
  * The payload format is identical to the nudge path so the skill receives
  * the same contract either way.
  */
@@ -465,7 +469,7 @@ export function claudeExtractionNudge(count: number, payload: string): string {
 
 /**
  * Static instructions for Cursor's AGENTS.md. Same contract as
- * claudeInstructions() — the host loads this at every session start, and the
+ * claudeInstructions() - the host loads this at every session start, and the
  * repo store name is auto-detected at runtime by the MCP server. Tool names
  * use the same mcp__thatch__ prefix since Cursor discovers MCP tools via the
  * same protocol.
@@ -507,18 +511,20 @@ you're working in. Prior sessions may have already investigated it.
 Thatch ships skills for code review, project investigation, and memory
 workflows. The host auto-discovers them, but reach for them proactively:
 
-- \`thatch-workflow-research\` — research code workflows affected by a change or planned change, before reviewing or planning.
-- \`thatch-change-walkthrough\` — produces a teaching walkthrough of a diff with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + change overlay mirroring the same numbers) and calibrated prose rules. Load the skill for the format and prose rules, not just for the research method.
-- \`thatch-code-walkthrough\` — produces a teaching walkthrough for a feature/workflow as it stands today with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + Key files) and calibrated prose rules. Also use proactively to draft high-level docs for new or undocumented features (suitable for pkg/README.md or docs/features/<name>.md). Load the skill for the format and prose rules, not just for the research method.
-- \`thatch-review-context\` — gather project context (PRs, tickets, TODOs, deferred work) before a review.
-- \`thatch-review-pedantic\` / \`-acceptance\` / \`-state-flow\` / \`-no-slop\` / \`-breadcrumbs\` / \`-mark-and-sweep\` / \`-highlights\` — seven specialist review lenses. Run individually, then \`thatch-review-synthesizer\` to verify and aggregate.
-- \`thatch-project-primer\` — investigate a new project and write foundational memories.
-- \`thatch-session-reflection\` — record what you learned at end of session.
-- \`thatch-fact-extractor\` — extract durable facts from recent tool interactions (auto-triggered by the extraction pipeline).
-- \`thatch-dedup-classifier\` — resolve duplicate memory pairs from \`find_duplicates\`.
-- \`pr-description\` — draft a PR description with SYNOPSIS / PURPOSE / DESCRIPTION / WALK-THROUGH / NOTES, using bold and italic emphasis on the phrases that carry the meaning.
-- \`ticket-description\` — draft a ticket, issue, or work item (Linear or Jira) with clear sections and bold/italic emphasis for scanning.
-- \`split-overlarge-pr\` — split already-completed work from an overlarge PR into human-reviewable, release-safe PRs targeting main.
+- \`thatch-code-archaeology\` - investigate an existing feature, debug an unfamiliar area, or begin a new ticket. Explores the code base from multiple angles (data model, state flow, git history, sibling features) before proposing changes. The research skill; pair with \`thatch-coding-workflow\`.
+- \`thatch-change-walkthrough\` - produces a teaching walkthrough of a diff with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + change overlay mirroring the same numbers) and calibrated prose rules. Load the skill for the format and prose rules, not just for the research method.
+- \`thatch-code-walkthrough\` - produces a teaching walkthrough for a feature/workflow as it stands today with a specific format (SYNOPSIS + per-workflow orient/mechanism/numbered-stages + Key files) and calibrated prose rules. Also use proactively to draft high-level docs for new or undocumented features (suitable for pkg/README.md or docs/features/<name>.md). Load the skill for the format and prose rules, not just for the research method.
+- \`thatch-review-context\` - gather project context (PRs, tickets, TODOs, deferred work) before a review.
+- \`thatch-review-followup\` - alternate entrypoint for follow-up review rounds. Verifies whether the author's responses and code changes since your last review adequately addressed prior findings, offers to reply on resolved items, then optionally re-runs the full structured review.
+- \`thatch-review-pedantic\` / \`-acceptance\` / \`-state-flow\` / \`-no-slop\` / \`-breadcrumbs\` / \`-mark-and-sweep\` / \`-highlights\` - seven specialist review lenses. Run individually, then \`thatch-review-synthesizer\` to verify and aggregate.
+- \`thatch-project-primer\` - investigate a new project and write foundational memories.
+- \`thatch-session-reflection\` - record what you learned at end of session.
+- \`thatch-fact-extractor\` - extract durable facts from recent tool interactions (auto-triggered by the extraction pipeline).
+- \`thatch-dedup-classifier\` - resolve duplicate memory pairs from \`find_duplicates\`.
+- \`thatch-pr-description\` - draft a PR description with SYNOPSIS / PURPOSE / DESCRIPTION / WALK-THROUGH / NOTES, using bold and italic emphasis on the phrases that carry the meaning.
+- \`thatch-ticket-description\` - draft a ticket, issue, or work item (Linear or Jira) with clear sections and bold/italic emphasis for scanning.
+- \`thatch-split-overlarge-pr\` - split already-completed work from an overlarge PR into human-reviewable, release-safe PRs targeting main.
+- \`thatch-review-response\` - respond to code review on the user's own PR. Triage findings, fix bugs one by one, reply on each thread, then post a top-level summary comment.
 
 ## When to Write
 
@@ -530,8 +536,8 @@ current state of the code).
 
 ## Before Responding
 
-Before composing a final response after substantial work this turn — multiple
-rounds of tool calls for investigation, debugging, or code-writing — check
+Before composing a final response after substantial work this turn - multiple
+rounds of tool calls for investigation, debugging, or code-writing - check
 whether you've discovered knowledge worth persisting. Use memory_recall to
 check for duplicates, then memory_remember for new findings. Then deliver
 your response.
@@ -601,7 +607,7 @@ do not save it. Do not store:
 
 ## Archived Memories
 
-Memories can be marked \`archived\` — a flag for stable, long-term historical
+Memories can be marked \`archived\` - a flag for stable, long-term historical
 records that should not trigger hygiene nudges (stale, orphaned, duplicate).
 Archived memories are excluded from search/recall results by default; pass
 \`includeArchived: true\` to surface them for archaeological dives.
@@ -609,23 +615,23 @@ Archived memories are excluded from search/recall results by default; pass
 When a branch is merged or about to be deleted, consolidate its branch-scoped
 memories into a single archived memory scoped to that same branch (preserving
 provenance). Capture intent, design decisions, review back-and-forth, PR
-number, unexpected pivots — the kind of git-archaeology context that explains
+number, unexpected pivots - the kind of git-archaeology context that explains
 _why_ the code looks the way it does a year from now. Then memory_forget the
 originals. The archived record outlives the branch. Future sessions searching
 with \`includeArchived: true\` can pull it up when investigating ambiguous code.
 
-Updating an archived memory requires passing \`archived\` explicitly — omit it
+Updating an archived memory requires passing \`archived\` explicitly - omit it
 and the tool returns an error. Pass \`archived: true\` to keep it archived,
 \`archived: false\` to unarchive.
 
 ## Explicit Requests
 
-"Remember X" — save immediately.
-"Forget X" — \`memory_recall\` to find it, then \`memory_forget\`.`;
+"Remember X" - save immediately.
+"Forget X" - \`memory_recall\` to find it, then \`memory_forget\`.`;
 }
 
 // ---------------------------------------------------------------------------
-// Prompt-aware recall nudge — fires when the user's prompt semantically
+// Prompt-aware recall nudge - fires when the user's prompt semantically
 // matches existing memories. Injected via opencode's chat.message hook
 // (in-process, warm model) or via the sideband socket for Claude Code/Cursor
 // (flush-tools CLI subcommand).