refactor(skills): retrofit the corpus to the ADR-0020 context contract (#129)

Retrofits all 39 skills to ADR-0020's description/body context contract, then fixes what six rounds of independent review found in that retrofit — including four ways the hot gate itself failed open.

Closes #99, #107, #108, #110, #111, #114, #115, #120.

## The retrofit (waves 1-5)

| | Start | Now |
|---|---|---|
| Description FAILs (>400 chars) | 26 | **0** |
| Body FAILs (>900 words, body-only) | 9 | **0** |
| Dangling routing targets | 2 | **0** |
| `Kyberforge.CompositionNote` | 10 | **0** |
| Preload tax | 21,005 chars | **~10,500** |

Under the 12,000-char success criterion. Per-wave detail is on #99.

## The review fixes

**The gate failed open four ways, three of them found after the retrofit shipped.** An unrecognised follower token made a dangling target vanish. A skill directory with no `SKILL.md` resolved as a valid target, so a commit could be green locally and red in a fresh clone — three existing fixtures were relying on that, one of which made the install-leak A/B pass vacuously. Then the free-standing `/name` sweep turned out to be gated on the sentence carrying a boundary marker, so route notation in any other sentence was invisible — not an ERROR, not a SUGGESTION, not an INFO — which left the documented "`/name` always blocks" promise false from a second direction. All four fixed and pinned.

**Two checks were silently not running.** `validate-provenance.sh` checks 7-8 were dead across nine skills. Waking them exposed a deeper problem: they assume `Research doc:` names a source index, but 30 of 121 entries point at topic content documents, so every new check-7 INFO was a false positive and check 8 was saved from a false-FAIL flood only by an *unannounced* skip. Checks 7/8 are now scoped to source indexes and every skip announces itself (#121).

**The retrofit's own anti-goal, four times.** ADR-0020 warns that a blunt gate gets satisfied by deleting content rather than relocating it. `diagnose` and `skill-audit` relocated prose and then read it unconditionally; `prototype` and `vale-config` deleted rules outright that survived nowhere. All four addressed.

## Verification

- `bash tests/run-tests.sh --strict` — 24 suites, 0 skipped, 0 failed
- `bash tests/run-bats.sh` — 325 tests, 0 failures
- `pre-commit run --all-files` — 17/17
- `pre-commit run --hook-stage pre-push --all-files` — 16/16, with `apm marketplace check` and `apm pack --check-clean` run against the remote, not skipped
- `scripts/skill-size-check.sh` over all 39 skills — rc 0, 0 ERROR/FAIL, SUGGESTION-only
- Preload tax measured at **10,498 chars**, max description 390 — both inside budget
- Every new test proven non-vacuous by a deliberate mutation of the behaviour it covers

**Per-commit sync, stated accurately:** the ten commits from the latest review round each pass `check-plugin-content-sync` in isolation, verified by checking each out in a detached worktree with a clean between. The earlier gitea window (`dfacf05..bedbd1d`, nine commits) does **not** — its mirror was regenerated in one batch at `bbc7300`. An earlier revision of this description claimed the property held for every commit; it does not, and a bisect through that window lands on a red commit. **Squash-merge** to collapse it, or accept that this range is not bisectable.

## Version bump

Six plugins and the catalog take a **patch**, not a minor. The branch is **89 commits — 40 `fix` / 30 `refactor` / 12 `docs` / 5 `chore` / 2 `test` — zero `feat`, zero `!`, zero `BREAKING CHANGE`** — and adds no skill, agent, command or hook. (Two earlier revisions of this section cited a stale histogram, most recently 78 commits; the figures above are measured at HEAD.) Both rules this repo ships (`forge/references/version-bump.md`, landing in this PR, and `git-commits/references/conventional-commits-spec.md`) make that a patch, and the catalog set is unchanged at 7 entries.

Not settled by that: four published files were removed from the installed tree, three moved, and `caveman` gained `disable-model-invocation`, retiring its old triggers. Under a strict reading those are major-class and currently ship under `refactor:` with no marker. Whether the deployed skill surface is a public contract is written down nowhere — worth deciding, but it outlives this PR.

## Deliberately not in scope

#112 (cherry-pick ownership, now resolved in favour of `git-commits`), #113 (`rtk git` normalisation), #116 (research fan-out), #101 (audit-skill merge), #122 (non-spec skill-root files), #123 (no PRD producer) stay open. #117 is the one worth reading: the contract's remedy is to move prose into `references/`, which is exactly where neither the size gate nor Vale looks — and the blind spot is wider than #117 currently records, since there is no root `.vale.ini` at all, so every ADR, `CONTEXT.md` and `README.md` is unlinted too.

That blind spot let this branch carry two `level: error` `Kyberforge.SentenceOpenerThereIs` violations into `references/` files it created — `provider-adapter-author/references/provider-matrix.md:31` and `agent-audit/references/finding-criteria.md:95`. Both are reworded in `afadaae`, confirmed by routing each file through the audit's own `vale-wrap.sh` (1 error each before, 0 after). Five further occurrences sit in `references/` files already on `main`; those are the pre-existing corpus and stay with #117, which is the real fix.

Also unfixed and not this PR's: `apm install` appends a duplicate `SessionStart` entry to `.claude/settings.json`, so a fresh clone cannot get pre-push green without an edit AGENTS.md warns against. Reproduces identically on `main`.

Co-authored-by: Defame1297 <gitea@rkdr.net>
Reviewed-on: https://git.dev.rkdr.net/Defame1297/holocron/pulls/129
Co-authored-by: Claude Code AI - Gitea MCP <claude@noreply.git.dev.rkdr.net>
Co-committed-by: Claude Code AI - Gitea MCP <claude@noreply.git.dev.rkdr.net>

plugins/kyberforge/skills/forge/references/apm-routes.md

@@ -0,0 +1,42 @@
+---
+source_keys:
+  - claude-code-subagents-docs
+---
+
+# Routing a plugin or marketplace entry to apm-workflow
+
+Reached from `SKILL.md` Step 2 when the classified artifact is a plugin or a marketplace entry.
+Both route to `apm-workflow` — a plugin to its configure flow (`apm plugin init`), a marketplace
+entry to its marketplace flow (`apm marketplace package add`).
+
+No other skill is a candidate for these two rows: `plugin-author` and `marketplace-author` were
+removed per ADR-0015 once issue #90 landed, and `apm-workflow` is their sole successor.
+
+## Always inline, never forked
+
+Run these routes inline, in the current conversation. Their flows are short, prompt-heavy or
+gated — `apm-workflow`'s publish and release steps take a HITL gate, and removing a marketplace
+entry takes a conversational confirmation — and a backgrounded fork cannot surface those
+checkpoints to the user in real time.
+
+## No clean-context recheck, and no automatic audit
+
+Skill and agent routes close with a clean-context audit rerun; these two do not, and the omission
+is deliberate rather than an oversight. Neither artifact type has an audit skill counterpart to
+re-run, so detaching the route to earn a recheck it would never get buys nothing.
+
+These routes get no automated terminal check either. `apm audit` is a separate action on
+`apm-workflow`'s own dispatch table, not a closing step of the configure or marketplace flow a
+forge route lands in, so a completion message from either says nothing about it. Do not wait for
+one and do not report one you did not see.
+
+Verify by hand instead. Read back what the route wrote against what the grill settled:
+
+- **Plugin** — the package directory exists where the intent said it should, and its `apm.yml`
+  carries the intended `name`, a top-level `type:` field, and a `version`.
+- **Marketplace entry** — the entry names that package, points at the source the intent settled
+  on, and carries the version the package actually declares.
+
+If the change warrants the full integrity and policy check rather than a read-back, invoke
+`apm-workflow` again for its audit action and run `apm audit` deliberately. Then return to
+`SKILL.md` Step 3 for the closing gates common to every route.

plugins/kyberforge/skills/forge/references/author-routes.md

@@ -0,0 +1,44 @@
+---
+source_keys:
+  - claude-code-subagents-docs
+---
+
+# Routing a skill or agent to its author skill
+
+Reached from `SKILL.md` Step 2 when the classified artifact is a skill or an agent/subagent
+definition. Route a skill to `skill-author` and an agent to `agent-author`. The two branches
+differ on one axis only — which audit skill verifies the result — and everything below applies to
+both.
+
+## Choose fork or inline
+
+Default to a **fork subagent**. It inherits the full grilled-intent conversation, so the author
+skill does not need re-briefing on what the user asked for or what the grill settled.
+
+Fall back to an **inline invocation** — same conversation, no subagent — when either holds:
+
+- **Fork is technically unavailable.** You are already running inside a fork (a fork cannot spawn
+  another fork), a nesting-depth cap is reached, or the environment does not support forking.
+- **The routed flow needs live user interaction mid-run** that a backgrounded fork cannot surface
+  in real time: clarifying questions, confirmation checkpoints, or a HITL gate. Judge this from
+  context — if nothing about the flow signals a live checkpoint, prefer the fork.
+
+## Two-tier verification
+
+Both author skills already close out with their own inline audit, in the same context as the
+authoring work: `skill-author` runs `/skill-audit`, `agent-author` invokes
+`agent-audit`. That is tier one, and forge does not change it.
+
+Tier two belongs to forge. Once the author skill's run has finished, spin up a separate
+**clean-context subagent** — fresh, not forked, no inherited context — to independently re-run the
+same audit skill against the finished artifact. This is a distinct verification layer, not a
+duplicate: the inline audit shares context with the work it is checking and can share its blind
+spots, while the clean rerun has no stake in the result.
+
+If the clean audit surfaces any unresolved finding — not only a disagreement with the inline pass,
+any actionable finding on its own — loop: re-invoke the author skill (same fork-versus-inline
+judgment as the first invocation) to resolve it, then re-run the clean audit. Repeat until the
+clean audit comes back with nothing unresolved. Only then is the route done. This is the same
+resolve-before-close discipline the author skills already apply to their own inline audit.
+
+Return to `SKILL.md` Step 3 for the closing gates common to every route once the loop closes.

plugins/kyberforge/skills/forge/references/sources.md

@@ -4,15 +4,15 @@
 
 - **URL:** https://code.claude.com/docs/en/sub-agents
 - **Research doc:** plugins/kyberforge/docs/research/docs/claude-code-plugins/sources.md
-- **Description:** Official Claude Code subagent reference — definition format, all frontmatter fields, scope priority, built-in agents, CLI flags, environment variables, known limitations. Grounds Step 3's fork-vs-inline invocation logic: fork inherits full conversation history via `/fork` or `subagent_type: "fork"`, is not a declarable frontmatter field on any agent definition, cannot be nested (a fork cannot spawn another fork), and is a caller-side invocation choice rather than a property of the artifact being routed to.
-- **Contributing files:** SKILL.md
+- **Description:** Official Claude Code subagent reference — definition format, all frontmatter fields, scope priority, built-in agents, CLI flags, environment variables, known limitations. Grounds the fork-vs-inline invocation logic in `references/author-routes.md`, the always-inline decision for the apm routes in `references/apm-routes.md`, and the clean-context bump subagent in `references/version-bump.md`: fork inherits full conversation history via `/fork` or `subagent_type: "fork"`, is not a declarable frontmatter field on any agent definition, cannot be nested (a fork cannot spawn another fork), and is a caller-side invocation choice rather than a property of the artifact being routed to.
+- **Contributing files:** SKILL.md, references/author-routes.md, references/apm-routes.md, references/version-bump.md
 - **Status:** `extracted`
 
 ## context7-websites-code-claude
 
 - **URL:** context7:/websites/code_claude
 - **Research doc:** plugins/kyberforge/docs/research/docs/claude-code-plugins/sources.md
-- **Description:** Official Claude Code documentation site indexed by Context7 — confirms the `context: fork` skill-level frontmatter field means isolated/fresh execution, the opposite of what the `/fork` subagent command does (inherits conversation). Informs the Gotchas entry warning against conflating the two.
+- **Description:** Official Claude Code documentation site indexed by Context7 — confirms the `context: fork` skill-level frontmatter field means isolated/fresh execution, the opposite of what the `/fork` subagent command does (inherits conversation). Informs the Gotchas entry in `SKILL.md` warning against conflating the two; nothing else in this skill draws on it, and no `references/` file mentions the `context: fork` field.
 - **Contributing files:** SKILL.md
 - **Status:** `extracted`
 
@@ -28,7 +28,7 @@
 
 - **URL:** https://agentskills.io/specification.md
 - **Research doc:** plugins/kyberforge/docs/research/docs/agentskillsio/sources.md
-- **Description:** Complete SKILL.md format specification — confirms `assets/`, `references/`, and `scripts/` are warranted only by the bulk/reusability of supporting content (large reference material, executable code, templates), not by a skill's category. forge has none of that bulk, so a lean SKILL.md-plus-provenance-file shape is spec-legitimate; the `references/sources.md` in this directory exists for this repo's own provenance-chain convention (see `CONTEXT.md`), not because the spec requires it.
+- **Description:** Complete SKILL.md format specification — confirms `assets/`, `references/`, and `scripts/` are warranted only by the bulk/reusability of supporting content (large reference material, executable code, templates), not by a skill's category, and forge's per-route procedures are that kind of supporting content — so the spec permits the `references/` split here but does not require it. The warrant is a house decision: ADR-0020's rule that dispatch is mandatory at two or more mutually exclusive flows, which forge's four-row table is. `references/sources.md` likewise exists for this repo's own provenance-chain convention (see `CONTEXT.md`), not because the spec requires it.
 - **Contributing files:** SKILL.md
 - **Status:** `extracted`
 

plugins/kyberforge/skills/forge/references/version-bump.md

@@ -0,0 +1,40 @@
+---
+source_keys:
+  - claude-code-subagents-docs
+---
+
+# Bumping the package version after a route
+
+Reached from `SKILL.md` Step 3 after a route has finished. A skill route always lands here:
+`skill-author` moves only a skill's own `metadata.version`, which is not the package manifest's
+number, so the package version is still behind when it reports done. `agent-author` bumps the
+resolved package's `apm.yml` itself at plugin/APM scope, and `apm-workflow`'s configure flow
+carries the same policy — read those routes' output before acting here, because a second bump for
+one change is wrong.
+
+## Find the owning package
+
+Walk up from the artifact's path to the nearest ancestor `apm.yml` that declares a top-level
+`type:` field (`instructions`, `skill`, `hybrid` or `prompts`).
+
+An `apm.yml` with **no** `type:` field is a marketplace-only manifest: it lists packages rather
+than declaring one, so it does not count as a match. Skip it and keep walking up.
+
+Skip this step entirely if no ancestor `apm.yml` carries a `type:` field: the artifact is then
+standalone or scoped to a user agent directory, and there is no package to version.
+
+## Delegate the bump
+
+Invoke `apm-workflow` as a **clean-context subagent** — fresh, not forked — with this
+brief:
+
+> "The package at `<package-path>` gained a new `<artifact-type>` (`<artifact-name>`). Bump the
+> `version` field in that package's `apm.yml`. Determine whether to bump minor (0.1.0) or patch
+> (0.0.1) based on whether this is a new capability (minor) or a fix/refactor (patch). Do not
+> release or tag — just update `apm.yml` and commit."
+
+Clean context rather than a fork is the point: the bump decision is made independently, without
+anchoring on the authoring conversation that just argued for the artifact's significance.
+
+Then report to the user: "Updated `<package-name>` version from `<old>` to `<new>` to reflect the
+new `<artifact-name>`."