Skip to content

chore: position idstack as Claude Code only (v3.8.0.0) - #121

Open
savvides wants to merge 1 commit into
fix/claude-code-shell-breakagesfrom
chore/claude-code-only-positioning
Open

savvides wants to merge 1 commit into
fix/claude-code-shell-breakagesfrom
chore/claude-code-only-positioning

Conversation

@savvides

Copy link
Copy Markdown
Owner

Stacked on #120. Merge that first, then this PR retargets to main.

What

  • One clear statement, two places. README (under the Requirement line) and the landing page install card now say: idstack supports Claude Code only; without it, use the Chrome extension; you can fork the repo and adapt the skills for another agent (e.g. Codex or Antigravity), but that's unsupported. Both lines carry IDSTACK_CLI_LEAK_ALLOW and are phrased as an unsupported-fork pointer, never as a support claim.
  • Landing badge "CLI / Multi-Agent Pipeline" → "Full 11-Skill Pipeline".
  • Multi-host residue removed: ROADMAP.md ("neither CLI", "graceful degradation elsewhere", "multi-platform support"), docs/why-ai-native.md ("LLM runtime" → Claude Code), README skill table, two test comments.
  • Agent hedges. The preamble and accessibility-review / course-builder / course-quality-review no longer branch on "if the Agent tool is not available" (it's always available in the main thread). They fall back only when a sub-agent fails. red-team keeps its sequential fallback, because a sub-agent can't start sub-agents.
  • Skill descriptions each get a "Use when …" trigger so Claude Code routes to them, use /idstack: names, and learn names idstack (it collided with another plugin's learn).

Guards

  • Leak sweep regex: codex|gemini → codex|gemini|antigravity.
  • The bare-/skill check now covers frontmatter descriptions (they were exempt); each description must contain "Use when".
  • Three mutation cases for these. mutation-test.sh: 151 guarded, 0 not guarded.

Verified locally

smoke-test (434, all pass except one local-only leak hit from a git-ignored nested clone that CI never sees), integration 51/51, setup 19/19, doctor 14/14, status 24/24, manifest-merge, version-classifier, plugin-status, preamble-python, skill-shell-zsh 47/47, extension, rendered landing (11 widths), check-doc-accuracy, check-citation-tiers, idstack-ste-check on the changelog entry.

Dropped from the audit plan

"Update check is dead for plugin installs": the only documented install is ./setup from a clone, which Claude Code loads from the clone (with .git), so the check and its git pull && ./setup hint already work.

🤖 Generated with Claude Code

- README and landing page: "Not using Claude Code?" note pointing to the
  Chrome extension; forks for other agents are unsupported. Landing badge
  "CLI / Multi-Agent Pipeline" -> "Full 11-Skill Pipeline".
- Remove multi-host residue in ROADMAP, docs/why-ai-native.md, README skill
  table and two test comments.
- Preamble and three skills drop "if the Agent tool is not available"
  fallbacks (it is always there in the main thread); they fall back only
  when a sub-agent fails. red-team keeps its sequential fallback because
  sub-agents cannot start sub-agents.
- Every skill description gets a "Use when ..." trigger and the namespaced
  /idstack: names; learn's description names idstack to avoid colliding with
  other plugins' learn skill.
- smoke-test: leak sweep adds "antigravity"; bare /skill check now covers
  frontmatter; each description must say "Use when". Mutation cases for all
  three (151 guarded, 0 not guarded).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant