@deftai/directive-content 0.77.0 → 0.78.0

package/skills/deft-directive-review-cycle/SKILL.md

@@ -1,9 +1,11 @@
 ---
 name: deft-directive-review-cycle
 description: >-
-  Greptile bot reviewer response workflow. Use when running a review cycle on
-  a PR -- to audit process prerequisites, fetch bot findings, fix all issues
-  in a single batch commit, and exit cleanly when no P0 or P1 issues remain.
+  Greptile / bot reviewer response and PR-shepherding workflow. Use when
+  running a review cycle or babysitting a PR to merge-ready -- audit process
+  prerequisites, fetch bot findings, fix all issues in a single batch commit,
+  and exit cleanly when no P0 or P1 issues remain. Supersedes Cursor global
+  babysit on Deft-managed repos (#2261).
 ---
 <!-- AUTO-GENERATED by task packs:render -- DO NOT EDIT MANUALLY -->
 <!-- Purpose: rendered skill -->
@@ -36,8 +38,17 @@ Legend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY.
 ## When to Use
 
 - User says "review cycle", "check reviews", or "run review cycle" on a PR
+- Operator asks to **babysit**, **shepherd**, or **watch** a PR -- including the Cursor product action **babysit-pull-request-in-cloud** (#2261 / #1862 intent-routing class)
 - A bot reviewer (Greptile) has posted findings on an open PR
-- Dispatching a cloud agent to monitor and resolve PR review findings
+- Dispatching a cloud or background agent to monitor and resolve PR review findings until merge-ready
+
+## Cursor global babysit supersession (#2261)
+
+On repos with `.deft/core/` installed, **PR shepherding intent** routes here -- not the Cursor-global `babysit` skill (`~/.cursor/skills-cursor/babysit/SKILL.md`). The product action **babysit-pull-request-in-cloud** shares the babysit name but MUST chain into this review-cycle surface on Deft-managed projects.
+
+- ! Orchestrators dispatching cloud/background workers for PR shepherding MUST include `templates/agent-prompt-preamble.md` (or reference it) and mandate reading this skill as the first workflow step.
+- ! Cloud babysit workers use `worker_role: review-monitor` (or `drive-to: merge-ready` implementation workers that own the review cycle) -- not a generic babysit loop without Phase 1 audit or Step 6 exit predicate (#1259).
+- ⊗ Route **babysit-pull-request-in-cloud** to Cursor global babysit when `.deft/core/` is present -- that bypasses xBRIEF preflight, `task check`, Greptile fail-closed exit, and CHANGELOG discipline (#2261 recurrence).
 
 ## Branch-Protection Policy Guard
 
@@ -389,6 +400,8 @@ task lifecycle:event -- emit plan:approved \
 
 ## Anti-Patterns
 
+- ⊗ Route PR shepherding to Cursor global `babysit` on Deft-managed repos when `.deft/core/` is installed -- use this review-cycle skill instead (#2261)
+- ⊗ Treat a passing SLizard/Greptile check run, a non-blocking review comment, or an ad hoc fix commit as the review-cycle exit predicate -- Step 6 fail-closed all-of (#1259) and multi-reviewer registry triage (#769) still apply
 - ⊗ Push individual fix commits per finding
 - ⊗ Start fixing before analyzing ALL findings
 - ⊗ Rely on the bot to catch syntax errors in structured data files

package/skills/deft-directive-setup/SKILL.md

@@ -482,7 +482,11 @@ omit = [
 
 - ~ After Phase 2 writes `PROJECT-DEFINITION.xbrief.json`, recommend `task triage:welcome` to the user as the single chained command for picking up the v0.27 triage surface. The N3 ritual (#1143) is the consolidating onboarding step for the #1119 governance swarm verbs (`task triage:bootstrap` / `task triage:scope` / `plan.policy.wipCap` writes / `task scope:demote --batch` relief / `task triage:summary`); without it consumers must learn each verb individually from the v0.27 release notes.
 - ~ `task triage:welcome` is idempotent and detection-bound -- each phase emits an informational stderr line and skips when its precondition is already satisfied, so a re-run after a partial completion resumes cleanly. The destructive phases (subscription / `wipCap` writes, optional WIP-relief invocation) are gated by numbered-menu prompts per [`../../contracts/deterministic-questions.md`](../../contracts/deterministic-questions.md). See [`../../UPGRADING.md`](../../UPGRADING.md) `## From v0.26.x -> v0.27` for the full walkthrough.
-- ? The recommendation is informational, not a hard gate -- consumers who plan to wire triage manually MAY skip the ritual and call the individual verbs in any order; the framework defaults stay fail-open per the umbrella `#1119 §12 framework-vs-consumer-config boundary`.
+### Consumer AGENTS always-pin recommendations (#2508)
+
+! After Phase 2 completes (or when closing a greenfield bootstrap), tell the consumer that process-critical directive skills should be named in their project's AGENTS.md **unmanaged header** when false-negative risk matters — full tier definitions: [`../../docs/skill-pin-policy.md`](../../docs/skill-pin-policy.md).
+~ Recommend the default consumer pin set for lifecycle work: `deft-directive-build` (implementation), `deft-directive-pre-pr` (before PR), `deft-directive-review-cycle` (bot review loop), and `deft-directive-swarm` when they plan parallel agents. Each is referenced by skill id + `.deft/core/.agents/skills/<id>/SKILL.md` path — not by pasting skill bodies.
+⊗ Recommend pinning entire `languages/`, `deployments/`, or other framework doc packs into AGENTS.md (#2508 anti-pattern).
 
 ---
 

package/skills/deft-directive-swarm/SKILL.md

@@ -57,6 +57,8 @@ This path became first-class in #1342 (platform adapter slices 1-3) and is fully
 
 ~ **Windows + Grok Build (#1353):** When issuing shell commands via `run_terminal_command` on this platform, avoid `|`, `>`, or `2>&1` in the command string — use Python `pathlib`/`subprocess` or plain `task` targets instead to avoid wrapper leakage. See `templates/agent-prompt-preamble.md` §3.5 for the full escape hatch list.
 
+! **Windows + Cursor Task-tool console storm (#2563):** Local Cursor Task subagents on Windows have frozen the host by flooding visible `cmd.exe` / `conhost` windows (observed with both parallel and serial Task agents; in-parent work stayed stable). Prefer one of: (1) in-parent / serial monitor-owned implementation, (2) cloud workers, or (3) after the #2563 mitigations (`windowsHide` + warm-dist skip in `engine:_ts-build`) are on the branch under test, keep local Task concurrency at **1** and prefer `pnpm exec` / warm `dist/bin.js` over repeated cold `task` rebuilds. See `templates/agent-prompt-preamble.md` §3.8. ⊗ Launch a wide parallel local Cursor Task cohort on Windows without that mitigation or an explicit operator acceptance of host-freeze risk.
+
 ## Prerequisites
 
 - ! `xbrief/active/` contains one or more story-level xBRIEFs with status `running`

package/tasks/engine-invoke.cjs

@@ -88,20 +88,23 @@ function main() {
     process.exit(2);
   }
 
+  // Command transport is one-hop: a spawned CLI may invoke Task again with a
+  // different ENGINE_CMD, which must not be shadowed by this inherited value.
+  const childEnv = { ...process.env };
+  delete childEnv.DEFT_ENGINE_CMD_JSON;
+  delete childEnv.DEFT_ENGINE_CMD;
+
+  // stdio inherit (not pipe): piped stdout/stderr deadlocks when the child emits
+  // more than the OS pipe buffer before exit — observed as greenfield smoke
+  // hanging then CI SIGTERM exit 143 with no output (#2554 / #2547).
   const result = spawnSync(execPath, execArgv, {
-    encoding: "utf8",
-    stdio: ["ignore", "pipe", "pipe"],
-    env: process.env,
+    stdio: "inherit",
+    env: childEnv,
     // Global deft/directive on Windows are .cmd shims; shell:false cannot spawn them (#2415).
     shell: mode === "global" && process.platform === "win32",
-    maxBuffer: 16 * 1024 * 1024,
+    // CREATE_NO_WINDOW: hide console windows from Cursor Task / nested shells (#2563).
+    windowsHide: true,
   });
-  if (result.stdout) {
-    process.stdout.write(result.stdout);
-  }
-  if (result.stderr) {
-    process.stderr.write(result.stderr);
-  }
   const code = result.status;
   process.exit(code === null ? 1 : code);
 }

package/tasks/engine.yml

@@ -35,13 +35,16 @@ tasks:
           // Windows native Task often has no `sh` on PATH, so Corepack.cmd was
           // invisible after #2411. Prefer a direct spawn (POSIX / real binaries);
           // fall back to shell:true so PATHEXT resolves .cmd/.exe on win32.
+          // windowsHide (#2563): CREATE_NO_WINDOW so Cursor Task shells do not
+          // flood visible cmd.exe/conhost windows on every probe/build.
+          const spawnOpts=(extra)=>({stdio:'ignore',windowsHide:true,...extra});
           const hasCmd=(name)=>{
             try{
-              execFileSync(name,['--version'],{stdio:'ignore'});
+              execFileSync(name,['--version'],spawnOpts());
               return true;
             }catch{
               try{
-                execFileSync(name,['--version'],{stdio:'ignore',shell:true});
+                execFileSync(name,['--version'],spawnOpts({shell:true}));
                 return true;
               }catch{
                 return false;
@@ -49,7 +52,7 @@ tasks:
             }
           };
           const run=(cmd,args)=>{
-            execFileSync(cmd,args,{cwd:root,stdio:'inherit',shell:true});
+            execFileSync(cmd,args,{cwd:root,stdio:'inherit',shell:true,windowsHide:true});
           };
           const trySteps=(steps)=>{
             for(const [cmd,args] of steps){
@@ -98,6 +101,10 @@ tasks:
     # ships no packages/ tree (guard false -> no-op, #2126). Git-vendored-payload
     # repos may carry a stray packages/ under .deft/core without a root `build`
     # script — source-only presence must NOT run `pnpm run build` there.
+    #
+    # Warm-dist skip (#2563): when packages/cli/dist/bin.js is newer than
+    # packages/{cli,core,types} sources, skip rebuild. Override with
+    # DEFT_FORCE_TS_BUILD=1; force-skip with DEFT_SKIP_TS_BUILD=1.
     dir: '{{.USER_WORKING_DIR}}'
     cmds:
       - |
@@ -105,6 +112,9 @@ tasks:
         if [ -f "{{.DEFT_ROOT}}/packages/cli/package.json" ] \
            && [ -f "{{.DEFT_ROOT}}/package.json" ] \
            && node -e "const fs=require('fs');const j=JSON.parse(fs.readFileSync(process.argv[1],'utf8'));process.exit(j.scripts&&j.scripts.build?0:1)" "{{.DEFT_ROOT}}/package.json"; then
+          if node "{{.TASKFILE_DIR}}/ts-build-fresh.cjs" "{{.DEFT_ROOT}}"; then
+            exit 0
+          fi
           node -e "
             const {execFileSync}=require('child_process');
             const fs=require('fs');
@@ -116,13 +126,16 @@ tasks:
             // Windows native Task often has no `sh` on PATH, so Corepack.cmd was
             // invisible after #2411. Prefer a direct spawn (POSIX / real binaries);
             // fall back to shell:true so PATHEXT resolves .cmd/.exe on win32.
+            // windowsHide (#2563): CREATE_NO_WINDOW so Cursor Task shells do not
+            // flood visible cmd.exe/conhost windows on every probe/build.
+            const spawnOpts=(extra)=>({stdio:'ignore',windowsHide:true,...extra});
             const hasCmd=(name)=>{
               try{
-                execFileSync(name,['--version'],{stdio:'ignore'});
+                execFileSync(name,['--version'],spawnOpts());
                 return true;
               }catch{
                 try{
-                  execFileSync(name,['--version'],{stdio:'ignore',shell:true});
+                  execFileSync(name,['--version'],spawnOpts({shell:true}));
                   return true;
                 }catch{
                   return false;
@@ -130,12 +143,20 @@ tasks:
               }
             };
             const run=(cmd,args)=>{
-              execFileSync(cmd,args,{cwd:root,stdio:'inherit',shell:true});
+              execFileSync(cmd,args,{cwd:root,stdio:'inherit',shell:true,windowsHide:true});
+            };
+            const markWarm=()=>{
+              try{
+                const dist=root+'/packages/cli/dist';
+                fs.mkdirSync(dist,{recursive:true});
+                fs.writeFileSync(dist+'/.deft-ts-build-stamp',new Date().toISOString());
+              }catch{}
             };
             const trySteps=(steps)=>{
               for(const [cmd,args] of steps){
                 try{
                   run(cmd,args);
+                  markWarm();
                   process.exit(0);
                 }catch{
                   // fall through to Corepack / next resolver
@@ -145,6 +166,7 @@ tasks:
             const envPm=String(process.env.DEFT_PACKAGE_MANAGER||'').trim().toLowerCase();
             if(envPm==='npm'){
               run('npm',['run',script]);
+              markWarm();
               process.exit(0);
             }
             const pin=String(pkg.packageManager||'').trim();

package/tasks/eval.yml

@@ -30,3 +30,12 @@ tasks:
       - task: :engine:invoke
         vars:
           ENGINE_CMD: 'eval:report --project-root "{{.USER_WORKING_DIR}}" {{.CLI_ARGS}}'
+
+  triggers:
+    desc: "Trigger routing coverage for AGENTS.md / Skills Index (#1586). Offline skill-pi-trigger-eval compatible eval over evals/trigger-cases.jsonl vs REFERENCES.md. -- task eval:triggers [-- --json] [--project-root PATH]"
+    deps:
+      - task: :engine:_ts-build
+    cmds:
+      - task: :engine:invoke
+        vars:
+          ENGINE_CMD: 'eval:triggers --project-root "{{.USER_WORKING_DIR}}" {{.CLI_ARGS}}'

package/tasks/ts-build-fresh.cjs

@@ -0,0 +1,99 @@
+#!/usr/bin/env node
+"use strict";
+
+/**
+ * Decide whether engine:_ts-build can skip `pnpm run build` (#2563).
+ *
+ * Exit codes:
+ *   0 — dist is fresh enough (or DEFT_SKIP_TS_BUILD=1); skip rebuild
+ *   1 — rebuild required (missing dist, newer sources, or DEFT_FORCE_TS_BUILD=1)
+ *   2 — usage / config error
+ *
+ * Warm-dist skip avoids a full `tsc -b` (and its nested shell:true spawns) on
+ * every `task <verb>` when packages/cli/dist/bin.js is already up to date.
+ */
+
+const fs = require("node:fs");
+const path = require("node:path");
+
+const SKIP_DIRS = new Set(["node_modules", "dist", "coverage", ".git"]);
+
+function walkSourceFiles(dir, out) {
+  if (!fs.existsSync(dir)) return;
+  let entries;
+  try {
+    entries = fs.readdirSync(dir, { withFileTypes: true });
+  } catch {
+    return;
+  }
+  for (const ent of entries) {
+    const full = path.join(dir, ent.name);
+    if (ent.isDirectory()) {
+      if (SKIP_DIRS.has(ent.name)) continue;
+      walkSourceFiles(full, out);
+      continue;
+    }
+    if (ent.isFile() && /\.(?:ts|tsx|json)$/.test(ent.name) && !ent.name.endsWith(".d.ts")) {
+      out.push(full);
+    }
+  }
+}
+
+function main() {
+  const root = process.argv[2];
+  if (!root) {
+    console.error("deft: ts-build-fresh usage: ts-build-fresh.cjs <DEFT_ROOT>");
+    process.exit(2);
+  }
+  if (process.env.DEFT_FORCE_TS_BUILD === "1") {
+    process.exit(1);
+  }
+  if (process.env.DEFT_SKIP_TS_BUILD === "1") {
+    process.exit(0);
+  }
+
+  const distDir = path.join(root, "packages", "cli", "dist");
+  const bin = path.join(distDir, "bin.js");
+  const stamp = path.join(distDir, ".deft-ts-build-stamp");
+  if (!fs.existsSync(bin)) {
+    process.exit(1);
+  }
+  // Prefer the post-build stamp (#2563): `tsc -b` incremental often leaves
+  // bin.js mtime unchanged even after a successful build when sources are
+  // newer than a previous emit. The stamp is written only after a completed
+  // engine:_ts-build, so it is the authoritative "warm dist" marker.
+  const marker = fs.existsSync(stamp) ? stamp : bin;
+  const markerMtime = fs.statSync(marker).mtimeMs;
+  const candidates = [];
+
+  for (const pkg of ["cli", "core", "types"]) {
+    const pkgRoot = path.join(root, "packages", pkg);
+    if (!fs.existsSync(pkgRoot)) continue;
+    walkSourceFiles(path.join(pkgRoot, "src"), candidates);
+    for (const name of ["package.json", "tsconfig.json", "tsconfig.build.json"]) {
+      const fp = path.join(pkgRoot, name);
+      if (fs.existsSync(fp)) candidates.push(fp);
+    }
+  }
+  for (const name of ["package.json", "tsconfig.json", "tsconfig.base.json", "pnpm-workspace.yaml"]) {
+    const fp = path.join(root, name);
+    if (fs.existsSync(fp)) candidates.push(fp);
+  }
+
+  for (const file of candidates) {
+    try {
+      if (fs.statSync(file).mtimeMs > markerMtime) {
+        process.exit(1);
+      }
+    } catch {
+      process.exit(1);
+    }
+  }
+  process.exit(0);
+}
+
+if (require.main === module) {
+  main();
+}
+
+module.exports = { walkSourceFiles };

package/tasks/verify.yml

@@ -415,6 +415,16 @@ tasks:
         vars:
           ENGINE_CMD: 'verify:eval-health-relocation --project-root "{{.USER_WORKING_DIR}}" {{.CLI_ARGS}}'
 
+  eval-triggers-relocation:
+    desc: "Fail-closed eval:triggers gate when AGENTS.md / REFERENCES.md / trigger cases change (#1586). Skips (exit 0) when the diff does not touch trigger-routing homes. -- task verify:eval-triggers-relocation [-- --base-ref <ref> | --staged] [--quiet]"
+    dir: '{{.USER_WORKING_DIR}}'
+    deps:
+      - task: :engine:_ts-build
+    cmds:
+      - task: :engine:invoke
+        vars:
+          ENGINE_CMD: 'verify:eval-triggers-relocation --project-root "{{.USER_WORKING_DIR}}" {{.CLI_ARGS}}'
+
   agents-md-advisory:
     desc: "ADVISORY (never fail-closing) consumer AGENTS.md legibility signal (#2155). Counts the UNMANAGED (project-authored) region only -- the framework-owned managed section is excluded -- and compares it against the SOFT, operator-adjustable budget plan.policy.agentsMdAdvisory.unmanagedSoftMaxLines (generous default when unset). Consumer companion to the maintainer-only #645 ratchet; deliberately NOT a check:consumer dependency (advise->observe->enforce per #1419). Default posture ALWAYS exits 0; the opt-in --enforce flag promotes an over-budget region to a hard cap (exit 1). Raising the field is the documented, no-friction way to accept growth and silence the nudge."
     dir: '{{.USER_WORKING_DIR}}'

package/templates/agent-prompt-preamble.md

@@ -211,6 +211,18 @@ PYTEST_ADDOPTS="--basetemp=$(mktemp -d)/pt" task check
 
 A clean result under an isolated basetemp is attributable to your change, not to the ambient shared-`/tmp` race. Do NOT point `--basetemp` at a static path shared across workers -- that re-introduces the collision. Solo / single-run invocations on a private worktree do not require this, but it is harmless to apply unconditionally.
 
+## 3.8 Windows Cursor Task-tool console storm (#2563)
+
+On Windows, Cursor Task-tool local subagents can open a visible `cmd.exe` / `conhost` window per shell turn. Framework source checkouts amplify this: many `task <verb>` calls run `engine:_ts-build` → `pnpm`/`tsc` via `shell: true`. Parallel (and even serial) Task agents have frozen the maintainer host; completing the same stories in-parent (no Task subagent) stayed stable.
+
+**Directive rule for orchestrators on Windows:**
+
+- ! Prefer in-parent implementation or cloud workers for Cursor Task cohorts until the branch under test includes the #2563 mitigations (`windowsHide` on engine spawns + warm-dist skip via `tasks/ts-build-fresh.cjs`).
+- ! When local Task agents are required, keep concurrency at **1**, reuse a warm `packages/cli/dist/bin.js` (set `DEFT_SKIP_TS_BUILD=1` only when dist is known current), and avoid spawning nested Task agents from workers.
+- ⊗ Launch a multi-agent local Cursor Task swarm on Windows that multiplies cold `task` rebuilds without operator acceptance of host-freeze risk.
+
+Reference: issue #2563; swarm skill Platform Requirements; env scrub + stdio inherit for nested Task recursion (#2554 / #2438) are necessary but not sufficient alone.
+
 ## 4. pre-pr and review-cycle skills
 
 Before pushing any branch:
@@ -225,8 +237,55 @@ Anti-pattern: pushing without pre-pr and relying on Greptile to find issues. Tha
 The active host harness may expose its own review-labeled surfaces. On Cursor these are the `bugbot` and `security-review` Task **subagent types** and the `review-bugbot` / `review-security` **skills**; other harnesses may ship equivalents. A generic operator request to "review" / "get this reviewed" / "use sub-agents for reviews" must NOT be routed to those host-native tools as the review of record.
 
 - ! Route ALL review work through the canonical `skills/deft-directive-review-cycle/SKILL.md` surface. Map a generic review request to the review cycle **by intent**, not by literal keyword -- "review this", "get this reviewed", and "use sub-agents for reviews" all mean run `deft-directive-review-cycle` (extends the #1862 / #2261 intent-routing fix).
-- ~ Host review tools (Cursor `bugbot` / `security-review` subagent types, `review-bugbot` / `review-security` skills, or any future host equivalent) MAY be folded in as *advisory* finding sources INSIDE the review cycle -- the #2019 harness-aware-reviewer path -- with their findings batched alongside the Greptile / bot findings the cycle already processes.
-- ⊗ Substitute a host-native review subagent type or `review-*` skill for `deft-directive-review-cycle` as the review surface. The host tools are advisory inputs folded into the cycle, never a replacement for it. Reaching for them on a bare "review" request is the 3rd recurrence of the #1862 / #2261 wrong-review-surface class (see also #2019, #2018).
+- ! Map **PR shepherding intent** the same way: `babysit`, `babysit this PR`, `shepherd`, `watch the PR`, and the Cursor product action **babysit-pull-request-in-cloud** all mean run `deft-directive-review-cycle` on Deft-managed repos (`.deft/core/` installed) -- NOT the Cursor-global `babysit` skill (`~/.cursor/skills-cursor/babysit/SKILL.md`) (#2261).
+- ~ Host review tools (Cursor `babysit` / `bugbot` / `security-review` subagent types, `review-bugbot` / `review-security` skills, or any future host equivalent) MAY be folded in as *advisory* finding sources INSIDE the review cycle -- the #2019 harness-aware-reviewer path -- with their findings batched alongside the Greptile / bot findings the cycle already processes.
+- ⊗ Substitute a host-native review subagent type, Cursor global `babysit`, or `review-*` skill for `deft-directive-review-cycle` as the review surface. The host tools are advisory inputs folded into the cycle, never a replacement for it. Reaching for them on a bare "review" or "babysit" request is the #1862 / #2261 wrong-review-surface class (see also #2019, #2018).
+
+## 4.6 Cloud PR-shepherd dispatch -- review-monitor worked example (#2261)
+
+When an operator triggers **babysit-pull-request-in-cloud** (or equivalent PR-shepherding intent) on a Deft-managed repo, the orchestrator MUST dispatch a **review-monitor** worker with this preamble (or a reference to `templates/agent-prompt-preamble.md`) and an explicit mandate to read `skills/deft-directive-review-cycle/SKILL.md` before any PR mutation.
+
+Worked example (cloud background review-monitor on PR #1037):
+
+```markdown
+## Allocation context
+
+- dispatch_kind: solo
+- allocation_plan_id: null
+- batching_rationale: null
+- cohort_vbriefs: []
+- operator_approval_evidence: operator selected babysit-pull-request-in-cloud 2026-07-03
+
+## Worker metadata
+
+- dispatch_provider: cursor-cloud-agent
+- worker_role: review-monitor
+- selected_backend: cursor-cloud
+- routing_policy: null
+- resolved_model: null
+- model_source: harness-default explicit
+
+## Runtime and GitHub auth mode
+
+- runtime_mode: cloud-headless
+- github_auth_mode: injected-token
+
+## Unit of work
+
+drive-to: merge-ready on PR #1037 (repo: deftai/deftvisage, branch: fix/visage-repo-org-scoping)
+
+## Mandates
+
+1. First tool call: read AGENTS.md; confirm Deft alignment.
+2. Read `templates/agent-prompt-preamble.md` (binding) and `skills/deft-directive-review-cycle/SKILL.md` end-to-end -- NOT `~/.cursor/skills-cursor/babysit/SKILL.md`.
+3. Run review-cycle Phase 1 process audit before the fix loop; batch Phase 1 + Phase 2 fixes per review-cycle discipline.
+4. Poll terminal verdict via `task pr:watch -- <N>` when waiting on Greptile/SLizard (#1056).
+5. Exit only on review-cycle Step 6 fail-closed all-of (#1259) -- ad hoc SLizard P2 fixes without the exit predicate are insufficient.
+
+DONE: include PR URL, role review-monitor, and whether Step 6 CLEAN was reached.
+```
+
+Anti-pattern: dispatching `Task(environment=cloud)` with only the Cursor global babysit skill attached and no preamble / `deft-directive-review-cycle` path -- that is the #2261 recurrence class.
 
 ## 5. REST-by-default for read-only gh calls
 

package/templates/agents-entry.md

@@ -57,9 +57,15 @@ Legacy `vbrief/` read-accepted; `deft migrate:xbrief` for `xbrief/` (v0.6→v0.8
 
 ! **Skills Index** (Level-0) in `.deft/core/REFERENCES.md` — scan before improvising; read `SKILL.md` only on index match. `welcome` / `onboard triage` → `deft triage:welcome --onboard` (N3 / #1143); lessons → packs:slice.
 
+## Skill pin policy (#2508)
+
+! Process-critical skills with false-negative risk MUST be named in AGENTS.md (always-pin tier) — tier definitions: `.deft/core/docs/skill-pin-policy.md` (#2508).
+! **Default always-pins:** `deft-directive-build`, `deft-directive-pre-pr`, `deft-directive-review-cycle`, `deft-directive-swarm` — read each `SKILL.md` when that work type starts.
+⊗ Pin entire language packs, deployment docs, or framework bulk into AGENTS.md — pins are for false-negative-sensitive process gates only (#2508).
+
 ## Review-surface precedence (#2308)
 
-! Route review through `deft-directive-review-cycle` — `.deft/core/.agents/skills/deft-directive-review-cycle/SKILL.md`; host `bugbot` / `security-review` / `review-*` advisory-only (#2308).
+! Route PR shepherding / review work through `deft-directive-review-cycle` — `.deft/core/.agents/skills/deft-directive-review-cycle/SKILL.md`; host `babysit` / `bugbot` / `security-review` advisory-only (#2308 / #2261).
 
 ## Value feedback and attribution (#1709)
 
@@ -67,7 +73,7 @@ Legacy `vbrief/` read-accepted; `deft migrate:xbrief` for `xbrief/` (v0.6→v0.8
 
 ## Eval and framework health (#1703)
 
-! `deft eval:health` when orienting or after gate/policy changes (Tier 0; 4-hour debounce). Release: `deft eval:run` / `deft eval:report` (#1703).
+! `deft eval:health` when orienting or after gate/policy changes (Tier 0; 4-hour debounce). Release: `deft eval:run` / `deft eval:report`; skill routing: `deft eval:triggers` (#1586 / #1703).
 
 ## Branch policy & branch verification