task-pipeline-skill 1.79.1 → 1.81.1

package/CHANGELOG.md

@@ -1,5 +1,229 @@
 # Changelog
 
+## v1.81.1 — the anchors are derived, and a dormant plant stops reading as a pass
+
+> **Why 1.81.1 and not 1.81.0.** A `v1.81.0` tag was cut from outside this run, against a tree that did not yet carry the run stamp, and its release round failed on the release-gap check. Nothing published; npm never left 1.80.0. The tag ruleset forbids deletion, so `v1.81.0` is declared dead in `docs/evidence/retro.md` → *Releases that carry no stamp* and this version carries the payload.
+
+Board row **B-113**, filed 2026-08-22 and confirmed four times since: *negative-test
+anchors are pinned to literals that the releases they guard move, so a release
+disarms its own checks.* Every confirmation had the same cause and the same reason
+nobody noticed — the plants were disarmed by the repository getting **healthier**.
+The board re-derived its ages and `bd5`'s needle stopped existing; the first blind
+eval runs shipped and `pf1`'s guard went dormant; a release finally carried an honest
+run stamp and `gap1`'s precondition emptied. All four were found by the 35-minute
+negatives suite at release time, one release after the damage, and until it ran
+everything was green.
+
+**The census, measured rather than estimated.** `test/anchors.py` reads all 422
+negative self-tests out of the workflow and answers the row's own question — *is the
+number the plant WRITES, or the number it LOOKS FOR?* — from the AST rather than by
+grep. A needle is only what a plant reads out of a file, before it writes that path:
+provenance is tracked from `open(...).read()` through derived names, read-backs are
+tracked per path (a plant-wide "after the first write" rule exempts `pf1`'s second
+file, which is a genuine anchor), regex shape is stripped (`[0-9a-f]{7,40}` is a
+shape, `\u2192` in a raw string is an escape and not the year 2192), and a
+`validate.py | grep 'message'` is assertion 3 rather than a needle. Counted over the
+corpus at the commit this release branched from: **26 of 419 plants pinned a value a
+release can move.** The row's own count of 27 came from a grep that both over- and
+under-reported — it missed `bd5`'s single-digit `| 1 | **6** |` entirely — and this
+release's own first figure of 21 was wrong for the same reason one level in: the
+detector could not see a needle read through `enumerate()`, `pathlib` or a compiled
+pattern. The R-005 reader found that by fixture, and it is why the number moved twice
+before it was true.
+
+**25 of the 26 are derived now**, one declared. Each reads its anchor out of the tree at run
+time: the high-water mark by shape, the rule number the stage map binds, the seeded
+templates' open row and worked example, the carry-over ledgers' board pointers, the
+live board's rows, the criterion numbers, the caps, the container count and its date,
+the recipe fence. The live-board plants were the sharpest: `bd10` pinned `| B-008 |`
+and `tbl`/`tbl2` pinned `| B-015 |`, so all three would have stopped landing the day
+those rows closed. One literal stays and says so: `res15` pins `^13\.` because the
+criterion NUMBER is the *guard's* contract — `test/validate.py` locates the teardown
+criterion as `^13\.` and its refusal names 13 — and deriving it would prove some other
+criterion's guard fired instead.
+
+**Dormant is no longer counted as passing.** `test/negatives.py` folded every plant
+that printed `SKIP` into *all N guards provably reject their planted defect*: a check
+that could not construct its precondition, reported inside the claim that every check
+rejected one. Skips now have their own bucket, print with the reason they gave, and
+the aggregate says *N of M · K DORMANT, named above — not counted as passing*. In CI
+each plant is its own step and a green step carrying SKIP is invisible, so the half
+that is checkable everywhere is the declaration: every plant that can decline to run
+carries `# dormant-when: <the state it cannot construct, and what that costs>`, and
+all four now do.
+
+**The class is self-reporting, in the cheap gate rather than the 35-minute one.**
+`test/validate.py` refuses a plant whose needle pins a moving value unless it declares
+why, refuses a declaration that resolves to nothing the plant reads, refuses a skip
+branch with no dormancy declaration, and refuses an empty census — because "no anchors
+found" and "the parser matched nothing" are otherwise the same sentence. Four checks
+ride into the suite with it, each watched red before shipping, and `test/anchors_test.py`
+is **47 whole-workflow fixtures, 27 of them watched firing** — a number that itself moved
+three times as the reader found what the census could not see. **Eighteen of the 45 cases assert SILENCE, and 16 of
+those 18 are retractions** — the other two are positive controls (`a derived needle passes`,
+`a declared anchor is accepted`) (the other two of the 47 checks are the empty-corpus
+census and a source check, which assert neither) — a `{7,40}` quantifier read as a year, `\u2192` in a raw
+string read as 2192, a payload built from local strings, a `validate.py | grep` read as a
+needle, a plant echoing `B-008` into its own replacement, `json.dump(d, open(p, "w"))`
+unrecognised as a write, a comprehension variable outliving its comprehension, and an
+unquoted `grep` taking a filename for its pattern. The other 25 fire, and four of them
+name blind spots the R-005 reader found: `enumerate()`, `pathlib.Path.read_text()`,
+`re.compile` and a heredoc spelling.
+
+**The R-005 reader found five blocking defects the author's own fixtures did not, and
+this is the part worth keeping.** A dispatched independent reader wrote twenty
+fixtures against the census and nineteen of them slipped past it. Provenance was lost
+through `enumerate()`, and **two live plants — `wv1` and `wv2` — were still pinned to
+`B-005` behind that blind spot** while the census called the corpus clean; the same for
+`pathlib.Path(...).read_text()` (seven plants) and `re.compile` (one). `bd2` and `bd13`
+had collapsed onto one row, so the dangling-id guard's well-formed branch went unprobed
+and `bd13` no longer proved what its name claims. And three documents restated a census
+of 21 that the detector itself put at 24. All five are closed, each with its own
+fixture; the true figure moved **21 → 24 → 26** as the detector stopped being blind, and
+that is the whole argument for the rule the reader exists to serve.
+
+Consequently the census now reads provenance through builtin wrappers, `pathlib`,
+`with`-handles, compiled patterns, `json.load`/`yaml.safe_load` and their structured
+looks, `sed`, `awk`, an unquoted `grep` and a `subprocess` argv; it keys read-backs by
+path SOURCE TEXT so a computed path matches itself; it refuses a heredoc spelling it
+cannot read rather than reporting silence — while ignoring one quoted inside a payload,
+because a guard reads what would RUN; and it judges a declaration on characters **and**
+distinct words, because sixty dots satisfy a length floor.
+
+**Round two of the same reader found four more, and one of them is this row's own
+class for the third time.** `json.dump(d, open(p, "w"))` was not recorded as a write, so
+the own-read-back of every plant that writes that way was one string literal from being
+refused — 39 plants write that way and **none was flagged yet**, which is a latent false
+positive and is said as one — the write side had been extended to
+`Path().write_text()` and not to `json.dump`, two halves of one rule landing unmatched in
+one commit. A loop or comprehension variable kept its file provenance for the rest of the
+body, so an unrelated later local read as file text. Needles 891 → **832**, and most of that
+drop is a second defect: a comprehension was dispatched twice, so needles inside its `iter`
+and `ifs` were counted twice over. The size of that share is **not stated here** — two
+methods of counting it disagreed (50 and 88) and neither has a command written beside it,
+which is the same defect one level up. 891 shipped as a measurement in a ledger row.
+The no-needle disclosure explained 39 plants as *a JSON key, which raises rather than
+passing*, and by the classifier that now ships exactly **1 of those 39** was a JSON or
+dict key: 13 were shell-only, 11 read no file, 10 compared whole-file bytes and 4 lost
+provenance at a helper function — so the breakdown is computed now — **five**
+reasons at this tree, none of them a gloss, and the fifth (`provenance stopped at a helper
+function`) is the one the gloss was hiding. And the fixture count was **stated as 18/8 in one sentence
+and 35/20 in the next**, which is B-113 itself, third occurrence inside its own fix
+(21 → 24 → 26, then 18/8 → 44/25 → 47/27). Every number in this entry and in the ledger rows is
+now pasted from the command that prints it.
+
+**A pure module for the arithmetic nobody can reach.** The dormant branch cannot fire on
+a healthy machine — all four skip-capable plants ran and their guards fired when the
+reader drove them end to end, because the runner rebuilds `.git` in its snapshot and CI
+fetches tags. So the two declarations that claimed otherwise were corrected after
+measurement, and `verdict()` and `claim()` became pure functions with
+`test/runner_test.py` calling them directly: 14 cases, two of them the empty-set branch
+that used to print `PASS: K DORMANT` over nothing at all.
+
+**One defect this change surfaced in a neighbour, fixed in the same commit.**
+`CONTRIBUTING.md`'s invariant list cites the guard literal that enforces each entry, and
+`test/validate.py` looked for those literals in its own source only — so the first
+citation of an anchors-census refusal was reported as *an enforcement that does not
+exist* while the check printing it sat in the same suite. The corpus is now this file
+plus the sibling modules it imports, discovered rather than listed. Invariant **64**
+carries the rule.
+
+**The loop guard tripped, and the trip is the release's own best evidence.** The fixture
+count returned twice with the same stale figure — inside the paragraph that narrates a
+stale figure as the row's lesson — because each fix APPENDED a corrected sentence instead
+of correcting the one already there. `build.md:417` says a tripped guard is not another
+round: stop, name the two shapes, escalate to the layer that owns the conflict. The layer
+is canon 8, and the escalation is mechanical: **`anchor-census fixtures` and
+`anchor-census fixtures watched firing` are registered claim classes now**, computed from
+the fixture file, with a plant watching them fail. Registered classes 13 → 15. Three
+figures this entry cannot compute are **deleted rather than corrected** — the share of the
+891 that was double-counted, the number of plants narrowing by order, and one the reader
+retracted itself — because a number whose definition is not written beside it is the
+defect this whole release is about.
+
+Guards: 419 → **423** — the re-pinned needle, the undeclared dormancy and the
+declaration that resolves to nothing, plus a property check that the census prints its
+disclosure beside the verdict.
+
+## v1.80.0 — the doctrine pack: one home per rule, and the boundaries said out loud
+
+The largest doctrine reconciliation since the corpus was audited whole. The
+wave-2/3 audit read all ~717KB of shipped doctrine against itself and this
+release closes its deferred pack: every contradiction between a short-form
+gate and its long form, every count restated where a list already was the
+count, and the two closure protocols that had no boundary sentence between
+them.
+
+**Two closure protocols, one boundary.** A work-graph node closes through the
+three blind tiers and `graph.py certify` (ceiling 3); a prose-plan task closes
+through `build.md` §4.4's single reviewer (five-round cap). Both were true and
+nothing said which applied — a run could pick whichever was cheaper at the
+moment of closing. The boundary sentence now lives in `SKILL.md`'s routing
+row, `certification.md` and `build.md`: the artifact the queue is decides.
+
+**The COPY/VISUAL track doctrine moved into `spec.md` whole** — its one home —
+and `stages.md`'s stage 3 keeps a routed summary; the example config's stage-3
+check now demands the tracks and their convergence record. `references/
+probing.md` is new: the four probe-authoring sections extracted from
+`gates.md` (plant/run/restore, how a probe rots, the neighbour probe, the
+landed-mutation rule), routed from the gate doctrine they prove; the guards
+and plants that read those sections moved with them, and the bundle's
+reference-file counts were re-derived to 38 by the registry that refuses a
+stale one.
+
+**The `judgment` gate type reaches every surface that taught `auto`/`manual`**
+— SKILL.md (three places), stages.md, the command, the Cursor rule, README,
+plugin.json — and a new cross-surface guard compares the schema's enum against
+the sentence that teaches it, so the next type cannot sit seven releases in
+the schema alone. The browser gate cells say when the look binds and what the
+weaker claim is; brainstorm's GATE names the queue-artifact and loop-arming
+criteria; the loop arms where `run.loop.arm` says, not unconditionally at
+preflight; stage-0 seeding covers the three gate scripts stages 5–10 run;
+the template board's State vocabulary admits `parked`/`waived`; project-audit
+prices with the board header's declared formula and never effort, and its
+disambiguation table gains the fifth audit (`/seo-aeo-audit`). Four domain
+companions (sheleg-dev, agent-stack, telegram-dev, seo-aeo-audit) enter the
+matrix, the preflight block and the stages they bind — recommended, never a
+gate. `model-tiering.md` gains the per-generation stress-test rule and the
+judgment gate the rubric-steering note, both with dated sources.
+
+**Mechanics: the PR gate now runs what `npm test` documents.** `graph_test.py`
+and `project_audit_test.py` reached CI only at tag time; a PR breaking
+`graph.py` merged green. The blind rule covers every prose field of a tier
+report — a cross-tier citation hiding in `scope` or a finding's `what`/`fix`
+is refused, fixtured, and mutation-proven (17/17 noticed, control 22).
+`gate-observer.sh` reads the documented `tool_response` payload with
+`tool_output` as fallback — the exit-code branch had been dead on the
+documented shape — with fixtures on both shapes, each watched red against the
+reverted hook. The upstream strict validation is its own CI job, so an npm
+outage and a house failure stop masking each other.
+
+**Ledgers: the board re-ranks on real ages.** Every open row's Age and Prio
+re-derived against its Source date — rows filed 08-17/08-20 still said Age 0
+on 08-29 under a header claiming re-derivation. The verification ledger's
+`Shipped state` header is gated against `package.json` (it had sat six
+releases stale), the two 2026-08-22 run stamps stop claiming retro entries
+that do not exist, the two rows frozen at `unreleased` gain superseding rows
+naming v1.72.0, and v1.76.0's absence from the ledger is stated beside the
+08-20 note. Five board rows filed from the harness-engineering articles
+(B-114…B-117) and from this release's own probes (B-118).
+
+**Evals: three blind runs, one per model.** evidence-docs and project-audit
+gain behavioural suites with near-miss negatives («аудит модуля» one word
+from «аудит проекта»); `evals/run.py` discovers every suite instead of the
+one filename it shipped pinned to; and 147 fresh-subagent probes (49 queries
+× haiku/sonnet/opus) produced the first dated blind rows in both RESULTS
+files — opus 42/42, sonnet 40/42, haiku 39/42, with haiku's three misses
+sharing one shape that is now B-118. Method and limits stated in the ledgers,
+INSTR cases recorded as not reproducible rather than guessed.
+
+`compatibility` frontmatter on task-pipeline and project-audit; a 3-line
+Degradation block in SKILL.md paid for by the model blockquote it displaced;
+`$schema` in both manifests.
+
+Guards: 417 → **419** — the gate-type vocabulary plant and the stale
+shipped-state-header plant, both watched red locally before shipping.
+
 ## v1.79.1 — the probe that a healthy release disarmed
 
 The first round of `v1.79.0` failed in the release job on the

package/CONTRIBUTING.md

@@ -592,6 +592,21 @@ file about itself. Checked in the live retro and in `templates/retro.md`, which
 false bound into every host project.
 *(guard: `declares the bound` and `and nothing enforces it`)*
 
+**64. A negative self-test may not promise that the tree will not change, and a plant that
+declines to run is not a plant that passed.** B-113, confirmed four times: a needle pinned to a
+value a release moves stops landing, and reports green while it does — and three of the four
+instances were caused by the repository getting *healthier* (a re-derived board, the first blind
+eval runs, the first honest run stamp), so nothing about them looked like a regression. Every
+needle a plant reads off disk before writing that path is extracted from the AST by
+`test/anchors.py`; one carrying a value a release can move must be derived at run time or
+declared in the plant's own body with what would falsify the declaration. Separately, any check
+that can print `SKIP` — a negative self-test **or a property check** — declares the state it
+cannot construct and what that costs, and `test/negatives.py` reports it by name instead of
+counting it inside *all N guards provably reject* or *N property check(s) printed what they
+assert*. Where nothing that ran proved anything the runner reports INCONCLUSIVE, because a
+pass over an empty set is a refused measurement.
+*(guard: `is anchored on a value a release can move` and `can print SKIP and declares no` and `the anchor census found no negative self-tests`)*
+
 ## Adding or changing doctrine
 
 - **Change one idea per PR.** These files are read by agents under load; a PR that

package/README.md

@@ -587,7 +587,7 @@ must say so.
 ### Held to Anthropic's own Skill authoring guidance
 
 Audited against the four Agent Skills pages. Most of it already held — `name`
-13/64 chars, `description` inside 1024, `SKILL.md` 334/500 lines, all 37 references
+13/64 chars, `description` inside 1024, `SKILL.md` 334/500 lines, all 38 references
 linked **directly** from `SKILL.md`, 436 KB against a 30 MB ceiling. What did not,
 now does:
 
@@ -787,7 +787,7 @@ Stages 0→10 above are the plugin's **example** flow. It is a machine-readable
 written against a universal contract
 ([`pipeline.schema.json`](plugins/task-pipeline/skills/task-pipeline/pipeline.schema.json)):
 copy the example to `pipeline.json` in your repo and rewrite it with your own stages
-(any count), your own `skills[]`, and your own `auto`/`manual` gate types. The
+(any count), your own `skills[]`, and your own `auto`/`judgment`/`manual` gate types. The
 framework bakes in no fixed stage count and no opinion on which gates are manual.
 
 ```jsonc
@@ -932,7 +932,8 @@ recommendation, so you arm the whole run in one exchange. Detail:
 | [`references/artifacts.md`](plugins/task-pipeline/skills/task-pipeline/references/artifacts.md) | the canonical document layout each stage writes to |
 | [`references/conventions.md`](plugins/task-pipeline/skills/task-pipeline/references/conventions.md) | how stages 6–10 read the host project's `CLAUDE.md`, and how the documentation regime is detected |
 | [`references/documentation.md`](plugins/task-pipeline/skills/task-pipeline/references/documentation.md) | the doc system: the inventory, registers and ids, SSOT, the Doc Loop, supersede semantics, the propagation matrix, intent vs as-built |
-| [`references/gates.md`](plugins/task-pipeline/skills/task-pipeline/references/gates.md) | the three axes, the promotion ladder, gate anatomy, the probe recipe, ratchet floors, where a gate runs |
+| [`references/gates.md`](plugins/task-pipeline/skills/task-pipeline/references/gates.md) | the three axes, the promotion ladder, gate anatomy, ratchet floors, where a gate runs |
+| [`references/probing.md`](plugins/task-pipeline/skills/task-pipeline/references/probing.md) | the probe recipe — plant, run, restore; how a probe rots; the neighbour probe; the landed-mutation rule |
 | [`references/deploy-targets.md`](plugins/task-pipeline/skills/task-pipeline/references/deploy-targets.md) | stages 7–8: writing the runbook when there is none, the template, per-platform deploy and log verbs, the verification trio |
 | [`references/hooks.md`](plugins/task-pipeline/skills/task-pipeline/references/hooks.md) | the `PreToolUse` contract, the fail-open hazard, placement, and the Claude-Code-only limit |
 | [`references/knowledge-graph.md`](plugins/task-pipeline/skills/task-pipeline/references/knowledge-graph.md) | the code graph: install line, stage-0 reach queries, the stage-9 refresh, the graph↔docs divergence check |

package/SKILL-CARD.md

@@ -12,7 +12,7 @@ harmless.
 |---|---|
 | **Purpose** | Runs a substantial task through ten gated delivery stages — intake grill, docs study, brainstorm, spec, plan, subagent build, tests, lint/deploy, post-deploy, docs+registers, acceptance — refusing to advance until each gate passes |
 | **Owner** | ssheleg ([github.com/ssheleg/task-pipeline](https://github.com/ssheleg/task-pipeline)) |
-| **Version** | 1.79.1 |
+| **Version** | 1.81.1 |
 | **Surface** | Claude Code (filesystem skill + plugin) and the vercel `skills` CLI. **Not** uploaded to the Skills API; custom Skills do not sync across surfaces |
 | **Dependencies** | None required. Optional: `context7` (MCP), `figma` (MCP), super-ux, agent-sync, graphify, obsidian-wiki, and **one of two browser channels** — `playwright` (CLI or MCP) or `chrome-devtools` (MCP); either satisfies the browser step and neither is required. Every stage's doctrine ships in-repo; the one conditional requirement is super-ux for the stage-3 UX track on a user-facing task |
 | **Evaluation status** | Suite authored, 5 categories. One recorded run, **self-observed by the author**; **zero blind runs on zero of three models** — the split, and the numbers, live in [`evals/RESULTS.md`](evals/RESULTS.md) and are computed by `evals/run.py` |

package/cursor/rules/task-pipeline.mdc

@@ -499,7 +499,9 @@ type the run produced, then run the documentation gate and print its ratchet cou
 ## Gates: three axes, and how to build one that cannot lie
 
 **Axis A — the stage gate type.** `auto` = verify the check yourself and stop on
-fail; `manual` = wait for an explicit operator go. An auto gate never substitutes
+fail; `judgment` = a named judge rules where no deterministic check exists, and the
+ruling is recorded as judgement, never as a measurement; `manual` = wait for an
+explicit operator go. Neither of the first two ever substitutes
 for a required manual approval.
 
 **Axis B — where the rule is actually enforced**, and a rule climbs this ladder

package/evals/RESULTS.md

@@ -1,9 +1,10 @@
 # Evaluation results — task-pipeline
 
-**Status: the suite is authored. One run is recorded and it was self-observed by the
-author; no blind run has been made on any model.** Recorded this way rather than left
-blank, because an empty results file and an unrun suite look identical, and this
-repository's own doctrine calls that the failure — a skip is not a pass.
+**Status: three suites authored (task-pipeline, evidence-docs, project-audit);
+one self-observed run and, since 2026-08-31, three blind trigger runs — one per
+model (haiku, sonnet, opus).** The instruction-following cases remain unexecuted
+and say so per run, because they need a full interactive session — a skip is not
+a pass, and a guessed verdict is worse than a vacant row.
 
 **The numbers below are computed, not asserted.** `python3 evals/run.py` counts the
 suite and the dated run headings in this file. A value typed here that disagrees with
@@ -39,11 +40,11 @@ or it belongs in a check.
 
 | Metric | Value | Computed by | As of |
 |---|---|---|---|
-| Evals authored | 15 | `python3 evals/run.py` → `suite: N evals` | 2026-08-08 |
-| Categories covered | 5 of 5 | the suite's own `category` fields | 2026-08-08 |
-| Dated runs recorded | **1** | `python3 evals/run.py` → `recorded runs: N` | 2026-08-08 |
-| …of those, **blind** | **0** | run headings not marked `self-observed` | 2026-08-08 |
-| Models exercised blind | **0 of 3** | distinct models across blind runs | 2026-08-08 |
+| Evals authored | 46 (three suites: task-pipeline 28, evidence-docs 9, project-audit 9) | `python3 evals/run.py` → `suite: N evals` | 2026-08-31 |
+| Categories covered | 5 of 5, per suite | the suite's own `category` fields | 2026-08-31 |
+| Dated runs recorded | **4** | `python3 evals/run.py` → `recorded runs: N` | 2026-08-31 |
+| …of those, **blind** | **3** | run headings not marked `self-observed` | 2026-08-31 |
+| Models exercised blind | **3 of 3** | distinct models across blind runs | 2026-08-31 |
 
 **The last three rows are the honest state of this skill's behavioural evidence**, and
 the split matters more than the total: a run the author watched, knowing the expected
@@ -67,6 +68,208 @@ green structural suite is never read as "the skill is known to work".
 Newest first. Each entry states the model, the task, and how the evidence was
 obtained — a self-observed run and a blind one are not the same claim.
 
+**Method (wave-3 subagent protocol, and its limits).** One FRESH general-purpose
+subagent per query per model, launched from a Claude Code harness with the model
+pinned. Each received: the query verbatim; the family's 28 skill
+names-with-descriptions (built once from the members' `SKILL.md` frontmatters,
+including evidence-docs and project-audit from this plugin); the instruction to
+answer with one skill name from that list, or `none`. Blind: no expected answer
+was named. Limits, stated rather than implied: the harness's OWN installed-skill
+inventory is also visible to a subagent's system prompt — a real coexistence
+environment rather than a cleanroom, and four early sonnet probes answered from
+it without opening the provided list (one chose `build-dashboard`, a skill
+outside the family; recorded as the miss it is, and the remaining prompts were
+tightened to "choose only from that file's list"). A trigger `pass` means the
+model NAMED the skill; it does not prove a full session loaded and obeyed the
+doctrine — the INSTR cases need exactly that and are recorded as **not
+reproducible from this harness**, not guessed.
+
+## 2026-08-31 · haiku · blind trigger probes, subagent protocol
+
+**39 of 42 scored cases passed** across the three suites' trigger-shaped
+categories (should_trigger, should_not_trigger; ambiguous and coexistence
+recorded unscored). Three misses, one shape: checkup/setup/prod-check phrasings («что у нас накопилось непроверенного?», «проверь документацию проекта…», «проверь, всё ли живо в проде…») routed to `project-audit` instead of task-pipeline's own checkup/setup modes — the two skills now share the "what is unverified/true" ground and the cheaper model takes the newer, narrower description.
+
+task-pipeline suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `task-pipeline` |
+| TRIG-02 | pass | named `task-pipeline` |
+| TRIG-03 | pass | named `task-pipeline` |
+| NOTRIG-01 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `none` — did not route to the excluded skill |
+| AMB-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+| AMB-02 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `ux-flows` (ambiguous/coexistence — recorded, not scored) |
+| TRIG-04 | pass | named `task-pipeline` |
+| NOTRIG-04 | pass | answered `none` — did not route to the excluded skill |
+| TRIG-05 | fail | answered `project-audit` |
+| TRIG-06 | fail | answered `project-audit` |
+| NOTRIG-05 | pass | answered `none` — did not route to the excluded skill |
+| TRIG-07 | pass | named `task-pipeline` |
+| TRIG-08 | pass | named `task-pipeline` |
+| TRIG-09 | fail | answered `project-audit` |
+| TRIG-10 | pass | named `task-pipeline` |
+| NOTRIG-06 | pass | answered `seo-aeo-audit` — did not route to the excluded skill |
+| NOTRIG-07 | pass | answered `ux-audit` — did not route to the excluded skill |
+| NOTRIG-08 | pass | answered `make-skill` — did not route to the excluded skill |
+
+evidence-docs suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `evidence-docs` |
+| TRIG-02 | pass | named `evidence-docs` |
+| TRIG-03 | pass | named `evidence-docs` |
+| NOTRIG-01 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `none` — did not route to the excluded skill |
+| AMB-01 | observed | answered `evidence-docs` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+
+project-audit suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `project-audit` |
+| TRIG-02 | pass | named `project-audit` |
+| TRIG-03 | pass | named `project-audit` |
+| NOTRIG-01 | pass | answered `task-pipeline` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `task-pipeline` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `make-skill` — did not route to the excluded skill |
+| AMB-01 | observed | answered `project-audit` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+
+INSTR-01…07 (task-pipeline) and INSTR-01 (evidence-docs, project-audit): **not
+reproducible from this harness** — each needs a full interactive pipeline run,
+and a guessed verdict would be the exact substitution this file exists to refuse.
+
+## 2026-08-31 · sonnet · blind trigger probes, subagent protocol
+
+**40 of 42 scored cases passed** across the three suites' trigger-shaped
+categories (should_trigger, should_not_trigger; ambiguous and coexistence
+recorded unscored). Two misses: TRIG-03 (`build a support-agent dashboard…`) answered `build-dashboard` — a skill OUTSIDE the offered list, from the harness's own inventory, before the choose-only-from-list tightening; TRIG-10 («посмотри PR #24 и скажи, что там не так») answered `none` — the PR-review findings class did not reach sonnet from the description alone.
+
+task-pipeline suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `task-pipeline:task-pipeline` |
+| TRIG-02 | pass | named `task-pipeline` |
+| TRIG-03 | fail | answered `build-dashboard` |
+| NOTRIG-01 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `none` — did not route to the excluded skill |
+| AMB-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+| AMB-02 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `ux-flows` (ambiguous/coexistence — recorded, not scored) |
+| TRIG-04 | pass | named `task-pipeline` |
+| NOTRIG-04 | pass | answered `none` — did not route to the excluded skill |
+| TRIG-05 | pass | named `task-pipeline` |
+| TRIG-06 | pass | named `task-pipeline` |
+| NOTRIG-05 | pass | answered `none` — did not route to the excluded skill |
+| TRIG-07 | pass | named `task-pipeline` |
+| TRIG-08 | pass | named `task-pipeline` |
+| TRIG-09 | pass | named `task-pipeline` |
+| TRIG-10 | fail | answered `none` |
+| NOTRIG-06 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-07 | pass | answered `super-ux:ux-audit` — did not route to the excluded skill |
+| NOTRIG-08 | pass | answered `make-skill:skill-audit` — did not route to the excluded skill |
+
+evidence-docs suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `evidence-docs` |
+| TRIG-02 | pass | named `evidence-docs` |
+| TRIG-03 | pass | named `evidence-docs` |
+| NOTRIG-01 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `none` — did not route to the excluded skill |
+| AMB-01 | observed | answered `evidence-docs` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+
+project-audit suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `task-pipeline:project-audit` |
+| TRIG-02 | pass | named `task-pipeline:project-audit` |
+| TRIG-03 | pass | named `project-audit` |
+| NOTRIG-01 | pass | answered `task-pipeline` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `task-pipeline` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `make-skill:skill-audit` — did not route to the excluded skill |
+| AMB-01 | observed | answered `task-pipeline:project-audit` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+
+INSTR-01…07 (task-pipeline) and INSTR-01 (evidence-docs, project-audit): **not
+reproducible from this harness** — each needs a full interactive pipeline run,
+and a guessed verdict would be the exact substitution this file exists to refuse.
+
+## 2026-08-31 · opus · blind trigger probes, subagent protocol
+
+**42 of 42 scored cases passed** across the three suites' trigger-shaped
+categories (should_trigger, should_not_trigger; ambiguous and coexistence
+recorded unscored). No misses. All four audit near-misses routed away correctly (module audit → task-pipeline, skill standard → make-skill, landing-page traffic → seo-aeo-audit, UX conformance → ux-audit).
+
+task-pipeline suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `task-pipeline` |
+| TRIG-02 | pass | named `task-pipeline` |
+| TRIG-03 | pass | named `task-pipeline` |
+| NOTRIG-01 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `none` — did not route to the excluded skill |
+| AMB-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+| AMB-02 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `ux-flows` (ambiguous/coexistence — recorded, not scored) |
+| TRIG-04 | pass | named `task-pipeline` |
+| NOTRIG-04 | pass | answered `none` — did not route to the excluded skill |
+| TRIG-05 | pass | named `task-pipeline` |
+| TRIG-06 | pass | named `task-pipeline` |
+| NOTRIG-05 | pass | answered `none` — did not route to the excluded skill |
+| TRIG-07 | pass | named `task-pipeline` |
+| TRIG-08 | pass | named `task-pipeline` |
+| TRIG-09 | pass | named `task-pipeline` |
+| TRIG-10 | pass | named `task-pipeline` |
+| NOTRIG-06 | pass | answered `seo-aeo-audit` — did not route to the excluded skill |
+| NOTRIG-07 | pass | answered `ux-audit` — did not route to the excluded skill |
+| NOTRIG-08 | pass | answered `make-skill` — did not route to the excluded skill |
+
+evidence-docs suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `evidence-docs` |
+| TRIG-02 | pass | named `evidence-docs` |
+| TRIG-03 | pass | named `evidence-docs` |
+| NOTRIG-01 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `none` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `none` — did not route to the excluded skill |
+| AMB-01 | observed | answered `evidence-docs` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `task-pipeline` (ambiguous/coexistence — recorded, not scored) |
+
+project-audit suite:
+
+| id | verdict | what happened |
+|---|---|---|
+| TRIG-01 | pass | named `project-audit` |
+| TRIG-02 | pass | named `project-audit` |
+| TRIG-03 | pass | named `project-audit` |
+| NOTRIG-01 | pass | answered `task-pipeline` — did not route to the excluded skill |
+| NOTRIG-02 | pass | answered `task-pipeline` — did not route to the excluded skill |
+| NOTRIG-03 | pass | answered `make-skill` — did not route to the excluded skill |
+| AMB-01 | observed | answered `project-audit` (ambiguous/coexistence — recorded, not scored) |
+| COEX-01 | observed | answered `project-audit` (ambiguous/coexistence — recorded, not scored) |
+
+INSTR-01…07 (task-pipeline) and INSTR-01 (evidence-docs, project-audit): **not
+reproducible from this harness** — each needs a full interactive pipeline run,
+and a guessed verdict would be the exact substitution this file exists to refuse.
+
 ## 2026-08-03 · opus · self-observed, not a blind run
 
 **Scope of this evidence.** The author ran the skill on a real task