Conversation
…uthCore (#157) Third deepening from the improve-codebase-architecture skill (design-it-twice → Design #2). The grant-based access model was mirrored across shares.ts and harnessShares.ts (two ~700-line files) — grantForToken, assertOwned*, resolveRole, activeGrantForCaller all near-identical — and the mirror had already rotted once (the sharedLocked cleanup fix landed on only one side). New shareAuthCore.ts owns the ONE security machine: empty-token→null guard, isActiveGrant (single impl) on the token lookup AND the per-user scan, opaque "Not found" (never leak existence), owner-wins-first, the cross-entity fk gate, editor-short-circuit / viewer-floor / max-role-wins, the truthy-userId guard (an anonymous caller rides the token path only — never matches a grantedToUserId===undefined row), and a required-projection sharedHeader (an unredacted egress is unrepresentable). Each entity wires it with concretely- typed closures via makeShareAuth(descriptor) — zero casts on the fk gate. Entity POLICY stays in the wrappers: publicMessage / publicHarnessProjection redaction, authorizeConversationWrite (owner|editor) vs editSharedHarness (editor-only + the sharedLocked lock + field allowlist), mint/rotate/list payloads, the harness lock-cleanup tail. Exported signatures unchanged (resolveConversationRole / authorizeConversationWrite / isActiveGrant re- exported from shares.ts) so messages.ts and other importers need zero edits. Behaviour-preserving: the 48 existing share/role tests (shares 32 + harnessShares 16) pass with ZERO expectation changes. New shareAuthCore.test.ts (12 cases) pins the invariants directly, incl. the anonymous-privilege- escalation negative and the cross-entity gate. convex 231 passed, tsc + biome clean. HOLD merge until AFTER v1.0 is cut (live authorization code — keep the release clean; a human should read the final diff).
…d .env The gateway's SKILL.md backfill (skill_content.py) calls the GitHub API unauthenticated from one EC2 IP — 60 req/hr shared across every user. Write the SKILLS_GITHUB_TOKEN secret into the host .env on deploy (replace-on- deploy, so rotating the secret rotates the host; skipped when unset). The Convex-side skill imports get the same token via the Convex env directly.
) Importing a repo like mattpocock/skills failed with "no skills found": the repo nests skills as skills/<category>/<id>/SKILL.md, but discovery only matched a fixed set of top-level bases (skills, .claude/skills, .agents/skills) exactly, so anything one folder deeper was invisible. Content-fetching already resolved nested skills by leaf-dir name (Convex viewer + gateway backfill both do), so only discovery was broken. Backend (convex/skills.ts): - discoverRepoSkills finds every SKILL.md at ANY depth in one git/trees call, derives a display "category" by stripping a recognized skills-root prefix (skills/.claude/.agents/.github), and skips node_modules. Leaf id must be unique within the repo (it's the fetch key) — first wins. - New listRepoSkills action: discovers + fetches + caches each SKILL.md (so previews are instant) and returns them with category + description + the repo's AGENTS.md/CLAUDE.md, for a choose-what-you-want flow. - importSkillRepo (curated templates, add-all) now shares the same discovery, so it benefits from nested support too. A repo that resolves but has no skills now reports "no skills found", not "repo not found". Frontend: - New RepoImportDialog: a two-pane browse-and-select modal — left is the discovered skills grouped by category with per-skill / per-group / select-all checkboxes and an added-already state; right is a live SKILL.md preview (rendered markdown, served from the browse cache so it's instant). Footer lets you also pull in AGENTS.md / CLAUDE.md. This brings repo import to parity with `npx skills add` (pick individual skills, see their contents). - The pack editor's "Import a GitHub repo" now opens this dialog (Browse) instead of a blind "Import all". Tests: convex/skills.test.ts stubs a nested (mattpocock-shaped) tree and pins discovery-at-depth, category derivation, node_modules exclusion, URL normalization, caching, and the no-skills vs not-found messages. convex 239 passed; web build + biome clean.
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.
Promotes staging to main. Carries:
mattpocock/skills(and anyskills/<category>/<id>nesting) failing to import.b6b74a5— theSKILLS_GITHUB_TOKENbackend deploy step; already present in main via ci(deploy): install SKILLS_GITHUB_TOKEN as GITHUB_TOKEN in the backend .env #160's cherry-pick, so it merges as a no-op.All staging checks are green and both convex+web deployed cleanly to staging. On merge, the prod frontend (Convex + Cloudflare) and backend deploy.