agent-afk 5.75.1 → 5.75.3

package/dist/bundled-plugins/awa-bundled/skills/devils-advocate/SKILL.md

@@ -24,18 +24,19 @@ When a proposal — a plan, fix, decomposition, scoping, or named recommendation
 4. Flag `dissent = true` when ≥2 critics returned `strong` alternatives disagreeing with the recommendation — signals the synthesizer is overruling well-argued dissent, so confidence is low. Include a `dissent_note` summarizing the strongest counter-argument.
 
 **Wave 3.5 — Composition-boundary check (fires on convergence):**
-Critics that converge may all have evaluated the proposal in artifact-isolation — none read the boundaries where it composes with siblings. A convergent verdict reached in isolation can be confidently wrong (e.g., critics agree on a UI glyph asserting visual continuity, but none saw that parallel-branch flushes reorder it). When the synthesis recommendation is **convergent** — recommendation ≠ original with `dissent = false`, OR ≥2 critics returned the same alternative — dispatch ONE context-injection verifier (same research-agent base) BEFORE surfacing:
+Critics that converge may all have evaluated the proposal in artifact-isolation — none read the boundaries where it composes with siblings. A convergent verdict reached in isolation can be confidently wrong (e.g., critics agree on a UI glyph asserting visual continuity, but none saw that parallel-branch flushes reorder it). When the synthesis recommendation is **convergent** — recommendation ≠ original with ≥2 critics having returned the same alternative — dispatch ONE context-injection verifier (**`subagent_type: "research-agent"`** — Read/Grep/Glob/WebFetch only, no Edit/commit) BEFORE surfacing:
 1. Its job is NOT to re-evaluate the proposal in isolation. It reads the 3 nearest composition boundaries — upstream caller, downstream consumer, and the render/event/state pipeline that interleaves the proposal's target with siblings.
 2. For each boundary: does the recommendation survive when the boundary varies? Check **temporal interleaving** (can flushes / parallel branches / sibling completions reorder it?), **state threading** (does it assume a point-of-use state upstream can break?), **adjacency assumptions** (does it presume render-tree / scrollback / call-graph adjacency that isn't load-bearing under recomposition?).
-3. Returns `CONFIRMED` only if the recommendation survives all three; otherwise `OVERRIDE: <specific boundary condition that breaks it>`.
+3. Returns `CONFIRMED` only if the recommendation survives all three; otherwise `OVERRIDE: <specific boundary condition that breaks it>`. (These verdicts are internal to Wave 3.5 — distinct from shadow-verify's verifier verdict vocabulary.)
 
-Until the verifier returns `CONFIRMED`, the convergent recommendation is a **candidate**, not a recommendation. On `OVERRIDE`, fold the named condition into the matrix and re-rank.
+Until the verifier returns `CONFIRMED`, the convergent recommendation is a **candidate**, not a recommendation. On `OVERRIDE`, fold the named condition into the matrix and re-rank. **Cap:** if `OVERRIDE` recurs after 2 re-ranks, escalate the full composition failure to the user rather than cycling further — the matrix cannot resolve a boundary violation on its own.
 
-**Scope guard:** skip when the proposal is purely local with no composition surface, or is anchored to an external referent that survives independently of the system. Fires once per convergent verdict, not per critic.
+**Scope guard:** skip when the proposal is purely local with no composition surface, or is anchored to an external referent that survives independently of the system. Does not fire when `dissent = true` — that path surfaces the matrix directly; adding a Wave 3.5 gate on already-uncertain output adds friction without signal. Fires once per convergent verdict, not per critic.
 
 **Merge + surface:**
-- Recommendation = `original` → the proposal survived critique; proceed with it — **unless ≥2 critics converged on the same alternative** (the second convergence condition above), in which case run Wave 3.5 first and, on `OVERRIDE`, re-rank before acting.
-- Recommendation ≠ `original`, `dissent = false` → synthesis found a better path; run Wave 3.5, then surface the alternative with rationale (on `OVERRIDE`, re-rank first) before acting.
+- Recommendation = `original` → the proposal survived critique; proceed with it.
+- Recommendation ≠ `original`, `dissent = false`, ≥2 critics returned the same alternative → convergent path: run Wave 3.5, then surface the alternative with rationale (on `OVERRIDE`, re-rank first) before acting.
+- Recommendation ≠ `original`, `dissent = false`, only 1 critic backed the winner → no convergence to guard: surface the alternative with rationale directly (Wave 3.5 does not fire).
 - `dissent = true` → present the matrix to the user; do not act. Confidence is low.
 
 **When to invoke:**

package/dist/bundled-plugins/awa-bundled/skills/diagnose/SKILL.md

@@ -6,6 +6,6 @@ context: fork
 
 Gather context: read the failing test or bug description, relevant error output, and recent git changes. If no failing test exists yet, write a minimal reproducer test (or identify a concrete verification command) before proceeding — hypotheses need a pass/fail signal to validate against. Dispatch two sub-agents in parallel — one to search the codebase for code paths involved in the failure (`subagent_type: research-agent`, read-only), and one to check recent commits and diffs that could have introduced the regression (`subagent_type: general-purpose` — requires Bash for `git log`/`git diff`/`git show`). When both return, synthesize findings into 2–4 ranked hypotheses, each with a specific code location and proposed cause.
 
-For each hypothesis, first create a dedicated git worktree with the `worktree` tool (`action: "create"`, one per hypothesis), then dispatch a sub-agent with its `cwd` set to that worktree's absolute path. This is how afk isolates a sub-agent to its own tree — the `agent` tool has NO `isolation` parameter, so parallel agents that omit `cwd` all inherit the same working tree and would contaminate each other's speculative fixes and test runs, invalidating the pass/fail attribution. Each sub-agent applies a minimal speculative fix, runs the test or verification command, then runs the broader related test suite to check for regressions. Run all hypothesis-testing agents in parallel, each in its own worktree; after collecting a result, remove that worktree with the `worktree` tool (`action: "remove"`). Collect results: which fixes passed, which didn't, and any regressions surfaced by the broader suite.
+For each hypothesis, dispatch a sub-agent with `isolation: "worktree"` to apply a minimal speculative fix, run the test or verification command, and then run the broader related test suite to check for regressions. Run all hypothesis-testing agents in parallel. Collect results: which fixes passed, which didn't, and any regressions surfaced by the broader suite.
 
 Report the validated root cause (the hypothesis whose fix passed), the speculative fix diff, and regression status from the broader test run. If no hypothesis passes, synthesize what was learned and form a second round of hypotheses. If the user approves the fix, apply it to the main worktree.

package/dist/bundled-plugins/awa-bundled/skills/shadow-verify/SKILL.md

@@ -18,15 +18,18 @@ When a sub-agent (or wave) returns investigation findings, code-review conclusio
 - `CONFIRMED` → surface the claim as validated.
 - `REFUTED` → replace the claim with the verifier's corrected finding, annotated `[was: confident, now: refuted]`, and show it alongside the original with evidence. Do not act until the conflict is resolved.
 - `UNVERIFIABLE` → surface with a `[needs-human-review]` tag rather than passing it through silently.
-- `UNVERIFIED-COMPOSITION` / `UNVERIFIED-ECHO-CHAMBER` (from the composition-axis guard below) → surface with a `[needs-human-review]` tag naming the missed boundary; do not pass through as validated.
+
+*The two verdicts below are **not** emitted by individual verifiers — they are produced by the Composition-axis guard (defined below) and handled here:*
+- `UNVERIFIED-COMPOSITION` → surface with `[needs-human-review: composition boundary unchecked]`; do not act until a boundary read confirms or refutes the claim.
+- `UNVERIFIED-ECHO-CHAMBER` → surface with `[needs-human-review: echo-chamber suspected]`; require at least one verifier to re-derive from outside the cited artifact before acting.
 
 Bound the loop: at most 3 verification rounds per session. Claims still unresolved after 3 rounds are escalated to the user, never silently dropped.
 
 **Composition-axis guard (echo-chamber check):**
 A verifier that re-derives a claim by re-reading the *same* file/region the original sub-agent cited has confirmed the citation, not the claim — it can be blind to composition-boundary failures (temporal interleaving, state threading, render/event-pipeline ordering, scrollback/call-graph adjacency) that only manifest outside the artifact's boundary. Before accepting a `CONFIRMED`:
 1. Read each verifier's `evidence_base`.
-2. For any **artifact-internal `CONFIRMED`**, require one composition-boundary read (≥1 upstream caller + ≥1 downstream consumer, plus the pipeline that interleaves the artifact with siblings) before merging. If a missed boundary surfaces, downgrade to `UNVERIFIED-COMPOSITION` and tag `[needs-human-review]`.
-3. **Echo-chamber guard:** if ≥2 verifiers cite the *same* in-repo artifact as primary evidence with no external referent, flag `UNVERIFIED-ECHO-CHAMBER` regardless of verdict and require one verifier to read outside that artifact's boundary.
+2. For any **artifact-internal `CONFIRMED`**, require one composition-boundary read (≥1 upstream caller + ≥1 downstream consumer, plus the pipeline that interleaves the artifact with siblings) before merging. If a missed boundary surfaces, downgrade to `UNVERIFIED-COMPOSITION` and tag `[needs-human-review]`. (An artifact-internal `REFUTED` is intentionally exempt: a refutation already halts action under the Merge rule above, so its boundary-blindness cannot drive a wrong commit — the asymmetry is safe by construction.)
+3. **Echo-chamber guard:** if ≥2 verifiers cite the *same* in-repo artifact as primary evidence with no external referent, flag `UNVERIFIED-ECHO-CHAMBER` regardless of verdict and require one verifier to read outside that artifact's boundary. If the 3-round loop cap is already exhausted when this fires, escalate to the user as `UNVERIFIED-ECHO-CHAMBER [loop-cap-reached]` — do not dispatch a new round.
 
 **Scope guard:** skip the composition check when the claim cites an external referent (RFC, spec, threat model, upstream-API contract) that survives independently of the repo, or when the artifact is purely local with no composition surface. Runs once per artifact, not on every cite.