Release v2.30.0

Released 2026-07-04. Additive MINOR (the distributed release zip gains content). No new skills; catalog stays 68 skills (30 phase + 11 foundation + 12 utility + 15 tool), 6 sub-agents unchanged.

The short version

A 2026-07-04 audit of this project found that its least-engineered surfaces carried its most damaging defects. QUICKSTART.md’s first line said “67 shipped PM skills” and then listed sub-counts that summed to 68, in the same line. The curated release zip omitted the agents/ and hooks/ directories the dispatch skills need at runtime, while still shipping the internal maintainer-docs tree. utility-pm-skill-builder, the gate every new skill passes through, carried a hand-maintained inventory that could not see the 15-skill tool family at all.

None of these were hard to fix. What made them dangerous was that they had shipped green through CI. This release fixes every instance of the drift and, more importantly, closes the gate holes that let it happen, so the “provable quality” claim this project makes is no longer contradicted by its own first sentence. The same pass then swept a batch of smaller hygiene items that had been accumulating: heading spelling drift, weak early-cohort descriptions, a frozen tool family with no version history, and CI ergonomics gaps.

What changed

Count truth, everywhere, with a gate that proves it

Every stale count on the front-door surfaces is fixed: QUICKSTART.md, both site quickstart pages, README.md, and the three plugin manifest descriptions now agree with each other and with the real catalog (68 skills, 11 foundation, 6 sub-agents). The seven “95+ sample outputs” mentions in README.md become a future-proof “200+” floor instead of a number that was already stale by a factor of two.

The durable fix is scripts/check-count-phrases.mjs, a new gate that derives the truth from disk and catches phrase variants the existing check-count-consistency gate cannot see: “N shipped PM skills”, “N skill definitions”, “N+ sample outputs”, and the “N sub-agents” axis, which nothing previously checked at all (this is exactly how the project’s public GitHub description drifted to claiming 4 sub-agents when there are 6). It shipped advisory first; this release promotes it to enforcing now that the sweep is verified clean.

The release zip now ships what it needs to work

The curated ZIP distributed through releases and the skills CLI omitted agents/ (the sub-agent definitions) and hooks/ (the activation hooks), even though several dispatch skills reference them directly. It now stages both, and continues to exclude the 400+ file docs/internal/ maintainer tree. New CI assertions check both directions: the six agent files and hooks.json must be present, and docs/internal must not be. .gitattributes now marks the internal paths export-ignore, so GitHub’s automatic source archives omit them too.

The skill-builder can see the whole catalog

utility-pm-skill-builder’s gap-analysis step is supposed to check a new skill idea against everything that already exists. Its inventory table listed 10 foundation skills under an “(11)” header and never mentioned the tool family at all, 22% of the catalog invisible to the one skill whose job is preventing collisions. It now derives its inventory from skill-manifest.json at run time instead of hand-listing it, so it cannot go stale again.

Cross-client documentation catches up to reality

The sub-agent compatibility matrix adds pm-skill-router and its current stamp; the platform setup guide fixes “All 4 sub-agents” and adds a real Gemini CLI install section; the ecosystem/MCP comparison page fixes a stale skill count and replaces an inaccurate “tracks releases 1:1” MCP claim with the honest frozen-at-v2.9.2 status. Three sub-agent definition files had dead documentation-path citations left over from the site’s Astro migration; those now resolve. Both quickstarts gain a per-client “Verify It Worked” step, so a new user can confirm the install landed before they start building.

Skill content mesh: boundaries, headings, and descriptions

Four skills gained “When NOT to Use” sections, closing two one-way reciprocity gaps in the boundary-pointer mesh. A new Skeleton Canon in the site skill-authoring guide names the three sanctioned SKILL.md dialects and their exact heading spellings, replacing a “mirror the closest exemplar” standard that had let a fourth, unsanctioned dialect emerge; 11 skills’ headings are normalized to it, and a new advisory check-heading-canon.mjs gate flags future drift. Eight early-cohort descriptions (including define-opportunity-tree and discover-journey-map) are rewritten to the standard set by foundation-build-risk-review: what the skill produces, when to use it, and at least one sibling deflection.

Tool family declared stable; CI gets easier to run

The 15 tool-* skills had sat at version 0.1.0 with zero HISTORY entries across seven releases, reading as permanently experimental. They move to 1.0.0 with a first HISTORY row each, and the HISTORY-start convention is now documented. Root package.json gains a test script that runs the full unit suite, replacing a hand-maintained ~21-file list inside the CI workflow; both setup-node steps cache npm; a new opt-in pre-commit hook runs the fast local subset of the release gate. New validators are frozen to single-source Node scripts going forward (no more .sh + .ps1 pairs), and the trigger-eval roster moves out of code into a data file that now correctly lists all 31 fixture sets instead of a drifted 29.

Marketplace pin and internal-docs visibility

The in-repo marketplace ref moves from the rolling main branch to this release’s own tag, so installs resolve the exact tree that was tagged rather than whatever main happens to hold at install time. Separately, the strategy/backlog/planning tier and the _agent-context/ coordination tree move to the gitignored local tier (kept on disk, no longer tracked); runbooks, the skill-versioning policy, and the release-plans history stay public.

What this means for you

• If you use the skills: nothing changes. Every skill, command, and workflow behaves exactly as it did in v2.29.1.
• If you installed via the release ZIP: re-download it. This is the first ZIP that carries agents/ and hooks/; if a dispatch skill previously failed to find its sub-agent definition on a ZIP install, this fixes it.
• If you contribute: check-count-phrases.mjs is now enforcing; a stale count phrase on a front-door surface fails CI. New validators should be single Node scripts, not shell pairs. An opt-in pre-commit hook is available (git config core.hooksPath .githooks).

Upgrade

# Update path
/plugin marketplace update
/plugin install pm-skills@product-on-purpose

# Or re-download the release ZIP if you installed that way.

No action required beyond updating. If you reference the catalog in your own tooling, it remains 68 skills, 6 sub-agents.

What does NOT change in v2.30.0

• All skills’ instructions, templates, and behavior are unchanged (content-mesh edits are boundaries, headings, and descriptions, not method changes).
• Skill catalog (68), slash commands (11), and workflows (12) are unchanged.
• The doc-stack, frontmatter structure, and cross-client dispatch mechanism are unchanged.

Affected areas

Area	Change
scripts/	New check-count-phrases.mjs (enforcing) and check-heading-canon.mjs (advisory); trigger-eval roster extracted to scripts/trigger-eval-roster.yaml; a dual-shell equivalence smoke over a new fixture tree; four orphan scripts removed.
.github/workflows/	validate-plugin.yml gains zip-stage assertions; validation.yml caches npm, calls npm test, and wires the two new gates.
.gitattributes	New export-ignore entries for the internal-docs and agent-coordination paths.
scripts/build-release.sh / .ps1	Now stage agents/ and hooks/ in the release zip.
skills/	utility-pm-skill-builder derives its inventory; 4 skills gain When-NOT-to-Use sections; 11 skills’ headings normalized; 8 descriptions rewritten; 15 tool skills bumped to 1.0.0 with first HISTORY rows. Catalog unchanged at 68.
hooks/	phase-router.mjs honors its opt-out; guardrails.mjs’s fabricated-metric regex requires metric context.
README.md / QUICKSTART.md / site	Count drift fixed; sample-count phrasing future-proofed; per-client verify step added; Gemini CLI section added; two broken site links fixed.
.claude-plugin/ / .codex-plugin/	Version 2.29.1 to 2.30.0; descriptions refreshed; marketplace ref pinned to v2.30.0.
docs/internal/	Strategy/backlog/next tier and _agent-context/ move to the gitignored local tier; check-context-currency retired.