Skip to content

v2.25.0: Activation and Trust Layer (hooks + eval harness)

Choose a tag to compare

View all tags
@github-actions github-actions released this 03 Jun 18:45
· 119 commits to main since this release
v2.25.0
23e65da
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Activation and Trust Layer

v2.25.0 wires the existing 65 skills into Claude Code's platform rather than adding more skills. It ships the plugin's first three pieces of activation-and-trust machinery. No new skills (catalog stays 65; sub-agents stay 5). Additive minor.

What's new

  • Opt-in house-rule guardrails (F-43) - a PreToolUse hook that blocks em-dash and en-dash characters at write time, but only when you enable it. It is OFF by default; set guardrails: true in a gitignored .claude/pm-skills.local.md to turn it on. Placeholder and unsourced-metric checks warn (never block). Covers Write / Edit / MultiEdit / NotebookEdit / ExitPlanMode, is dependency-free, and fails open.
  • Confident-only phase router (F-44) - a SessionStart hook that suggests the right Triple Diamond skills for where you are, but only when a repo signal is strong (a phase-named branch, or a single unambiguous PM artifact). It stays silent otherwise, so it never becomes noise.
  • Output-quality eval harness (M-30) - three deterministic invariant validators over the recorded skill samples (no placeholders, exact-quote sourcing, no fabricated metrics), wired advisory in CI.

Hooks are a Claude Code primitive, so the guardrails and router are Claude Code features; the portable surface across other clients remains the skills.

Do I need to do anything?

No. The guardrails are off until you opt in, and the router only speaks when it is confident. Everything you already use behaves exactly as before.

Learn more

  • Concept guide: the new "Hooks and Output-Quality Checks" page (how the hooks work and the full .claude/pm-skills.local.md schema).
  • Full changelog: CHANGELOG.md

What's Included

  • PM skills library (flat skills/{phase-skill}/ + foundation capabilities)
  • Sample output library (library/skill-output-samples/ with README and canonical sample corpus)
  • Slash commands (10 /workflow-* orchestrators; every skill invocable by name)
  • Workflows (Triple Diamond, Lean Startup, Feature Kickoff)
  • Claude plugin manifest (.claude-plugin/plugin.json)
  • Claude/openskills discovery via sync helper (scripts/sync-claude.sh / .ps1)
  • Docs: README, QUICKSTART, AGENTS, reference guides

Download

pm-skills-v2.25.0.zip . Complete package for all platforms

Installation

Claude Code / openskills

  1. Download pm-skills-v2.25.0.zip
  2. Extract, then run:
    • Bash: ./scripts/sync-claude.sh
    • PowerShell: ./scripts/sync-claude.ps1
  3. Claude Code / openskills will discover commands in .claude/commands/ and skills in .claude/skills/.

Claude plugin clients

  1. Download pm-skills-v2.25.0.zip
  2. Extract the archive.
  3. Point plugin setup to .claude-plugin/plugin.json from the extracted package.

Other editors/agents

  • Use skills/ and commands/ directly from the extracted ZIP.

Quick Usage

/pm-skills:deliver-prd "Your feature description"
/pm-skills:define-hypothesis "Your assumption to test"
/workflow-feature-kickoff "Feature name"

See QUICKSTART.md in the archive for full instructions.

Assets 5
Loading