feat(marketplace): unify Firstsun External catalog across ChatGPT/Codex and Claude Code - #15
Merged
Merged
Conversation
Add the firstsun-external catalog (7 plugins, 86 skills) plus loadCatalog/discoverSkills library functions and a test suite covering field validation, source-root safety, in-plugin and cross-plugin duplicate-name rejection, expectedSkillCount enforcement, and a repository-level contract test against the real external/ tree. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Add buildOutputs/writeOutputs/checkOutputs to external-marketplace.mjs so the Codex (.agents/plugins/marketplace.json) and Claude Code (.claude-plugin/marketplace.json) manifests, plus the seven catalog plugin packages under plugins/, are generated purely from marketplace/external/catalog.json and the real skill files on disk -- deterministic, real files only (symlinks rejected), catalog order for plugins and alphabetical order for skills. sync-external-marketplace.mjs now supports --write and --check modes; --check is read-only and reports missing/changed/unexpected files as drift. Regenerate the repository outputs: replace the personal tianyao-skills/most-used-skills bundle and the five legacy external-* plugins with the seven canonical catalog plugins (agent-toolkit, software-delivery, spec-driven-development, frontend-product-design, visual-content, career-productivity, health-wellness), and retire plugin-most-used/. This is the intentional clean break described in the cross-platform external marketplace design. Verified: node scripts/test-external-marketplace.mjs passes (including all prior Task 1 tests), sync --write followed by sync --check exits 0 with no drift, and a second --write produces zero further git diff (byte-identical determinism). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
buildOutputs copied skill files via readFileSync/writeFileSync, which silently dropped the executable bit -- 19 scripts across agent-toolkit, software-delivery, and visual-content (e.g. brainstorming's start-server.sh, skill-creator's *.py helpers) regenerated as mode 644 instead of the original 755, breaking direct execution. Track each copied Buffer's source mode in a module-internal WeakMap and re-apply it with chmodSync in writeOutputs, without changing the public buildOutputs/writeOutputs Map<string, string | Buffer> contract. Add a test asserting a chmod 755 fixture script keeps its mode after writeOutputs. Regenerated plugins/ so all 19 previously-flagged files are 755 again; re-verified zero further diff on a second --write and that `sync-external-marketplace.mjs --check` still exits 0. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Wire node scripts/sync-external-marketplace.mjs --check into ./init.sh (new "external marketplace parity" gate) and into the weekly update-external-skills workflow as an explicit verification step after the refresh. Also broaden the workflow's add-paths to cover both platform manifests, the canonical catalog, and the full plugins/ tree (replacing the now-obsolete plugins/external-*/** glob). Add a test proving checkOutputs actually reports drift when a tracked marketplace output file is tampered with. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Rewrite README's marketplace section and SKILLS_LIST.md's Plugins table to describe the single firstsun-external catalog (7 plugins across Build/Design/Grow) published to both ChatGPT/Codex and Claude Code, and add a documentation contract test asserting the obsolete tianyao-skills, most-used-skills, and external-career-health names are gone and every plugin id, manifest path, and verified skill count is present. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
feat-016 done: canonical catalog generates matching Codex/Claude manifests across 7 plugins covering all 86 external skills, with parity enforced in init.sh/CI. Records final verification evidence in feature_list.json and closes out progress.md. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…t-check gap, stale comment, vacuous doc test) - Fix critical bug where the weekly sync workflow's cleanup step (`rm -rf .agents .claude/skills`) deleted the tracked .agents/plugins/marketplace.json, causing every run's ./init.sh step to fail on "missing: .agents/plugins/marketplace.json". Now only removes the actually-gitignored .agents/skills subdirectory. - Add a new verify-marketplace CI job (triggered on push/pull_request to main) that runs the contract tests and sync --check directly against committed output, closing the gap where a hand-edited generated file could sit undetected until the next weekly sync. - Update stale "stage these two" comment to reflect the current six-entry add-paths list. - Scope the SKILLS_LIST.md doc-count assertion to the line(s) describing each specific plugin instead of a whole-file substring search, so a drifted per-plugin count can actually fail the test. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
marketplace/external/catalog.json) that generates identicalfirstsun-externalmarketplace manifests for both ChatGPT/Codex (.agents/plugins/marketplace.json) and Claude Code (.claude-plugin/marketplace.json), organized into seven task-oriented plugins across the Build/Design/Grow taxonomy, covering all 86 external skills with zero overlap.plugins/<id>/) contains real copied files (no symlinks) under a sharedskills/directory with both.codex-plugin/plugin.jsonand.claude-plugin/plugin.jsonmanifests, so the package is portable when imported from GitHub.scripts/sync-external-marketplace.mjs --write/--checkis the single deterministic generator; a full test suite (scripts/test-external-marketplace.mjs) covers catalog validation, discovery, generation, and drift detection.init.shand adds a dedicated push/PR-triggered CI job (verify-marketplace) so a hand-edited generated file now actually fails CI instead of silently drifting until the next weekly sync.tianyao-skills/most-used-skillsbundle and the five legacyplugins/external-*packages from the public marketplace surface.README.mdandSKILLS_LIST.mdto document one catalog, two clients..agents/plugins/marketplace.json(only.agents/skills/is meant to be gitignored) — this branch's newinit.shparity gate would otherwise have made every weekly run fail.Implemented via Subagent-Driven Development: fresh implementer + task review per task, plus a final whole-branch review that caught the CI cleanup bug and the tautological drift check above (both fixed and re-reviewed clean).
Test plan
node scripts/test-external-marketplace.mjspassesnode scripts/sync-external-marketplace.mjs --checkexits 0 with no driftgit diff(deterministic)./init.shpasses end-to-end including the new external-marketplace-parity gatefind plugins -mindepth 1 -maxdepth 1 -type d | wc -l= 7;find plugins -mindepth 4 -maxdepth 4 -name SKILL.md | wc -l= 86;find plugins -type l | wc -l= 0git diff -- skills-lock.jsonempty (no provenance drift);git diff --stat -- custom/empty (custom/ untouched)verify-marketplaceCI job runs the test suite +--checkon push/PR without a preceding--write(not tautological)🤖 Generated with Claude Code