Skip to content

v5.27.0 — Skill Rationalization & /prune

Choose a tag to compare

View all tags
@samibs samibs released this 13 Jul 06:33
· 26 commits to main since this release
v5.27.0
780fbef
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Skill Rationalization & /prune

A new capability and a disciplined cleanup.

New: /prune — Dead Code & Duplication Remover

Finds and safely removes dead code and duplicate code with a real detection engine — not prompt guesswork.

  • Duplicate blocks (N-line window matching, cross-file), unused imports (JS/TS/Python) with a safe --fix, unused-export candidates (reported only).
  • Safe by default: only fully-unused imports are auto-removed; exports and duplicates are reported for review (an export may be an entry point; de-dup is a refactor). --deep runs jscpd/knip/depcheck when present.

Skill rationalization: 109 → 97 skills, zero capability lost

SkillFoundry had accumulated prompt-era scaffolding — redundant personas and thin "guardian" role cards built to force weaker models into rigor. Smarter models need fewer, sharper tools.

  • Removed migration-note "skills" (upgrade-*, agent-index), four *-guardian role cards, and the redundant secure-coder / security-guardian / pentest personas.
  • Kept or universalized the good content — red-team-researcher and security-scanner stay (they're genuinely valuable); the senior-engineer discipline (assumption-surfacing, anti-sycophancy push-back) is now in _coding-discipline.md, applied to every agent instead of one opt-in skill. Threshold rules from the guardians (≥95% coverage, load-test ≥1000 users, blast-radius) were preserved in _bpsbs-production-rules.md.

The guiding rule: remove scaffolding, never quality.

Full detail in the CHANGELOG and release notes. Full test suite: 198 passing / 0 failing.

Assets 2
Loading