Skip to content

v1.3.0

Latest
Latest

Choose a tag to compare

View all tags
@codybrom codybrom released this 22 Jul 16:35
· 3 commits to main since this release
v1.3.0
fcfda01
This commit was signed with the committer’s verified signature.
codybrom Cody Bromley
SSH Key Fingerprint: oP3Yy5mGCqHG/qM4ioV8yRczRb3dA3WfsoIL0DMhWp4
Verified
Learn about vigilant mode.

Removed

  • All hooks are gone — the plugin now ships only skills and the clean-room-alternative agent.

    • The keyword-suggestion hook (prompt-keywords.sh): field telemetry showed its per-skill grep over the full session transcript hitting the hook's 5s timeout on nearly every prompt in long or MCP-heavy sessions — a UserPromptSubmit hook blocks the agent loop, so this added ~5 seconds of latency to almost every prompt, and the practical user response was to disable the whole plugin.
    • The plan-mode nudge (enter-plan-mode.sh): its PostToolUse matcher targeted the EnterPlanMode tool, which only exists in newer Claude Code versions, and the agent doesn't need a hook to consider design alternatives when planning.

    Skill routing is served by the skill descriptions themselves and by /clairvoyance:diagnose for vague symptoms — the agent picks skills fine without prompt-time nudging.

  • The per-session state directory ~/.claude/clairvoyance/ is no longer used or written to. It was never cleaned up automatically, so if you ran earlier versions you can safely delete it.

Added

  • CI validation: a GitHub Actions workflow now runs claude plugin validate --strict, skill/agent frontmatter consistency checks (scripts/check-skills.sh), and a cross-manifest version-sync check on every PR touching skills or manifests — the class of check that would have caught 1.2.0's inert-metadata: frontmatter bug two releases earlier.
  • scripts/bump-version.sh: sets the version across all 8 manifests that carry one (Claude, Codex, Cursor, Kimi, Gemini, npm, docs) from a single declarative list, with a --check mode used in CI.
  • Marketplace entry metadata: category, tags, homepage, and repository for discoverability.

Changed

  • Tightened all 16 skill descriptions by ~20% (7,073 → 5,659 chars of context resident in every session's skill listing), keeping the "Use when…" triggers and "Not for X (use Y)" discriminators that drive routing while cutting duplicated clauses.

Fixed

  • Platform manifest versions (Codex, Cursor, Kimi, Gemini, docs site) had drifted to 1.2.0 while the Claude manifests said 1.3.0 — now synced via the new bump script.
  • docs: the OpenCode install guide's version-pinning example no longer shows a stale tag.
Assets 2
Loading