feat(skills): publish complete skill archives with hardened generation - #95
Merged
Conversation
Replace the per-file skill-md discovery entries with one complete tar.gz archive per skill, built from a single commit-pinned steel-dev/skills snapshot so the manifest and every supporting file come from the same commit. Archive bytes are deterministic (normalized tar headers and gzip metadata), digested with sha256, and the build fails closed on any catalog error. Hardening from code review: - Exclude hidden files/dirs from archives and cap uncompressed content at 5 MB per skill, so stray upstream files (.env.local, fixtures) can never be republished on docs.steel.dev - Fall back to the manifest key when a skill entry omits path, matching upstream's own `meta.path ?? name` resolution - Attribute YAML frontmatter parse failures to the offending skill - Exclude the whole RFC 8615 /.well-known namespace and archive extensions (.gz/.tar/.tgz) from markdown content negotiation - Serve index.json and tarballs with a short shared Cache-Control lifetime so a stale index cannot digest-mismatch fresh tarballs - Run agent-skills generation in CI so failures surface pre-deploy
Contributor
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
nibzard
added a commit
that referenced
this pull request
Jul 31, 2026
…ling the build (#108) PR #95 rewrote the generator and dropped the deploy-safe error handling from PR #89, so any GitHub outage, 5xx, or exhausted rate limit aborted the whole production build via bun run generate. Restore the intended two-tier split: - Transport errors (fetch network failures, non-ok GitHub responses, rate limits) are classified as GitHubTransportError. The entrypoint warns and exits 0, leaving the index absent for that deploy. - Schema, packaging, validation, and filesystem errors still throw and fail the build, so a malformed index can never ship. The script writes nothing before every artifact validates, so skipping on a transport error leaves no partial output behind.
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.
What
Follows up #89: the Agent Skills discovery index now publishes one complete tar.gz archive per skill (
type: "archive") instead of per-fileskill-mdentries, so agents fetch a skill's SKILL.md together with its references, scripts, and templates in a single download.Hardening (from code review)
.env.localor oversized fixture upstream cannot reach docs.steel.dev or OOM a deploy.pathis optional, falling back to the skill name exactly like upstream'smeta.path ?? name, so a legal upstream manifest cannot break docs deploys./.well-known(RFC 8615 namespace is machine-readable by definition) plus.gz/.tar/.tgzassets, replacing three one-off path entries.max-age=300, must-revalidate) onindex.jsonand the tarballs: they change together on deploy, so a stale index can no longer fetch tarballs whose digest it must reject as tampered.secrets.GITHUB_TOKEN, surfacing rate-limit or upstream drift failures in PRs instead of at deploy time.Testing
/.well-knownand archive-extension negotiation.35bf2783, none containing hidden entries.