Three related half-applied changes from #130, each leaving the corpus in a state its own
documentation contradicts.
Why:
- `assets/templates/SKILL.md` shipped `metadata:` fully commented out, and `new-skill.sh` only
substitutes SKILL_NAME. Every scaffolded skill therefore lacked the `metadata.version` ADR-0022
made mandatory and was blocked at first commit by the very hook this PR added. The commented
example also read `"1.0"` — neither the `0.1.0` new-skill seed nor valid semver.
- `agent-audit/references/scope-project-user.md` still joined `disable-model-invocation` and
`user-invocable` with a slash — #125's defect verbatim — while pointing the reader at the file
this PR had just corrected to say the opposite.
- ADR-0022 required the "when present" bump conditional dropped and `metadata.version` moved into
create.md's required list. It was dropped from SKILL.md but left in README.md, and the field was
edited in place under a heading that still authorises removing it entirely.
Implementation notes:
- The template emits `metadata: version: "0.1.0"` live, captioned as required, with the optional
keys left commented. `new-skill.bats` gains a case asserting a live key and three-part semver, so
this cannot regress.
- `description-quality.md` now asserts only what the vendored Copilot research supports: two fields
with opposite defaults, and the retired `infer` replaced by the pair rather than by either alone.
The unsupported negative it previously stated as fact is gone.
- The `1.0.0` retrofit seed is stated in improve.md and retrofit.md, which the retrofit flow
actually reads — create.md, where it lived, is unreachable from that path. The compression item
moved out of the file-churn checklist, whose preamble excluded the wording-only change it covers.
- Executable git commands in these three skills now carry the ADR-0023 rtk prefix.
Refs: #125, #127
ADR: 0022, 0023
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EeH8SCbcrCAQrtymkNuhKP
Diffing each retrofitted SKILL.md against its replacement references/ files found
rules that existed on main and now existed nowhere — relocated in intent, deleted in
fact. A trim that loses a rule is not progressive disclosure, it is data loss with a
smaller word count.
Three had no survivor. The least-privilege guidance for `tools` kept its mechanics
and lost the "restrict to what the agent needs" half, so the remaining text read as
encouragement to omit the field. The improve flow lost its regression check, so
nothing compared the closing audit against the pre-edit state and a PASS quietly
becoming a SUGGESTION went unnoticed — restored on both halves of the author pair,
since agent-author had dropped its equivalent too. And agent bodies lost "would the
agent get this wrong without it?", which mattered more than it looks: ADR-0020
deliberately sets no body word gate for agents, three of the four already sit
between 933 and 1,199 words, and the delegation check only fires on procedure a
skill already owns. That heuristic was the only brake left.
Two more were reachable only from the wrong scope. agent-author tells the reader to
load only the file for the resolved scope, but the mcp__ glob syntax for
disallowedTools and the five tools no subagent ever receives had both landed in
project-user-scope.md. disallowedTools is the ONLY permitted fence at plugin/APM
scope, so the scope that needs the syntax most could not reach it, and a plugin-scope
run could write a body telling the agent to ask the user a question.
Two documents were actively wrong rather than merely thin. agent-audit told auditors
that validate.sh resolves boundary targets for skills only; it runs at both scopes,
so the auditor was hand-resolving what the script had already decided and could
contradict it. And skill-audit routed to its script-troubleshooting reference
whenever validate.sh "fails" — but it exits 1 on ordinary content FAILs, the normal
outcome for the whole #99 population, so 1,302 words loaded on nearly every audit.
A context-budget regression inside the skill that enforces the context budget.
Finally, two illustrations taught the shape the gate ERRORs on, unfenced, while an
adjacent rubric called it a hard ERROR.
LESSONS.md records the reference-chain depth rule flipping from "one level deep" to
"two hops, never three". ADR-0020 is silent on it and the reversal rode entirely on
the diff; the looser rule is what mandatory dispatch requires.
Refs: #99
ADR: 0020
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015W3iwF9ncfRZddGBxsMCYi
ADR-0020 shipped its gates hot with no baseline file, so 26 of 39 descriptions
and 9 of 39 bodies are over their FAIL tier and editing any of them for any
reason requires bringing the skill into contract first. `references/improve.md`
said exactly that and stopped there — it mandated a retrofit and supplied no
procedure for one.
Four dry-run retrofits confirmed what that costs. Asked the same questions —
what to cut first, when a body is two flows rather than one, what else has to
change alongside — they invented six to ten different answers, so the same skill
retrofitted twice produced two different skills and neither run could be reviewed
against anything.
`references/retrofit.md` fixes the answers: an ordered cut list ranked by tokens
removed against behaviour lost (inverting that order is how a retrofit deletes the
instruction the skill existed to carry), the test for whether a body holds two
mutually exclusive flows, the reference-file conventions, the collateral checklist
for `README.md` and `references/sources.md`, and a worked description retrofit.
It also states the trap the dry runs kept hitting: retrofit the skill in place,
inside its package. The boundary-target universe is built by walking up from the
file being checked, so a scratch copy has no authoring root above it, the check
prints `INFO ... DID NOT RUN`, and the run still exits 0 — a line that reads as a
pass and is not one. A retrofit signed off on a copy carries an unverified
boundary target into the corpus.
Loaded from the improve flow only when a budget is actually exceeded, so a routine
improvement pays nothing for it.
Refs: ADR-0020, #99
Skill name+description pairs are preloaded into every session, costing
~6,200 tokens across 39 skills before any skill is invoked. The authoring
rules mandated that growth: skill-author:104 and description-quality.md:21
both required padding, while skill-author:102 (the deflating rule) had no
FAIL condition behind it.
Gates (blocking, no baseline file):
- description 250 chars SUGGESTION / 400 FAIL, measured on the folded
YAML value
- body-only 600 words SUGGESTION / 900 FAIL, independent of the unchanged
whole-file 2770-word / 500-line spec backstop
- every boundary-clause routing target must resolve to a real skill or
agent; catches skill-improve, neuledge-context and gitea-labels
- agents take the description gates but deliberately no body gate; a test
pins that absence
Vale: DescriptionOpener widened to ^This\b, new CompositionNote rule
banning architecture notes from descriptions. 10 hits, 0 false positives.
Kyberforge's own four skills retrofitted: descriptions 3,364 -> 938 chars
(-72%), bodies 8,306 -> 2,487 words (-70%), all via the apm-workflow
dispatch pattern. Fixes the skill-improve dangling route and the
agent-author misroute to manual review.
Also fixes a pre-existing false positive where any line-initial 'read '
was flagged as interactive input, which had already caused two scripts to
be rewritten around it.
Refs: ADR-0020
