SLSA Provenance Review Capability Pack Skill
Expert SLSA provenance review skill for checking source, build, artifact, and trust evidence before accepting content or package submissions.
Open the source and read safety notes before installing.
Safety notes
- Use this as a read-only evidence review workflow; ordinary intake should not require running submitted projects.
- The install command fetches a pinned SLSA source archive for reference, not an executable package.
- Provenance evidence improves confidence but does not replace human source review, category-fit review, or maintainer judgment.
- Treat missing, stale, unverifiable, or mismatched provenance as a risk signal and request clearer evidence before accepting strong trust claims.
Privacy notes
- Work from public source URLs, release metadata, package metadata, and maintainer-approved context.
- Keep public notes focused on source URLs, revisions, digests, and high-level risk summaries; omit operational details that do not need to be public.
Prerequisites
- Submission target such as a registry entry, package listing, source-backed content item, or release artifact
- Canonical source repository, docs page, release page, package URL, commit, tag, or digest evidence
- Expected producer, repository, source revision, artifact name, and artifact digest when available
- Access to provenance records, verification summaries, or package ecosystem metadata when the producer publishes them
- Local policy for deciding when missing provenance is acceptable, blocking, or manual-review only
Schema details
- Install type
- package
- Reading time
- 9 min
- Troubleshooting
- Yes
- Scope
- Source repo
- Skill type
- capability-pack
- Skill level
- expert
- Verification
- validated
- Verified at
- 2026-06-03
| Platform | Support | Install path |
|---|---|---|
| claude-code | Native | .claude/skills/<skill-name>/SKILL.md |
| codex | Native | .agents/skills/<skill-name>/SKILL.md |
| windsurf | Native | .windsurf/skills/<skill-name>/SKILL.md |
| gemini | Native | .gemini/skills/<skill-name>/SKILL.md or .agents/skills/<skill-name>/SKILL.md |
| cursor | Adapter | .cursor/rules/<skill-name>.mdc |
| cli | Manual | AGENTS.md or tool-specific context file |
Full copyable content
# Trigger
"Apply the SLSA provenance review capability pack to this submission."
# Required output
1) Source, package, artifact, and provenance evidence inventory
2) Expected producer, repository, revision, builder, and policy
3) SLSA source/build provenance review result
4) Acceptance, rejection, or manual-review decision with evidenceAbout this resource
Knowledge Freshness
This capability pack is pinned to SLSA v1.2 documentation and source files verified on 2026-06-03. Refresh the SLSA spec, provenance format, and package ecosystem behavior before using it for production policy decisions.
Retrieval Sources
- https://slsa.dev/spec/v1.2/provenance
- https://slsa.dev/spec/v1.2/verifying-artifacts
- https://slsa.dev/spec/v1.2/verifying-source
- https://slsa.dev/spec/v1.2/build-provenance
- https://slsa.dev/spec/v1.2/source-requirements
- https://slsa.dev/spec/v1.2/distributing-provenance
- https://slsa.dev/spec/v1.2/threats
- https://raw.githubusercontent.com/slsa-framework/slsa/v1.2/docs/spec/v1.2/provenance.md
- https://raw.githubusercontent.com/slsa-framework/slsa/v1.2/docs/spec/v1.2/verifying-artifacts.md
- https://raw.githubusercontent.com/slsa-framework/slsa/v1.2/docs/spec/v1.2/verifying-source.md
Prefer current SLSA specification pages and pinned source files over model memory for predicate details, verification expectations, and level-specific requirements.
Scope Note
This is not a HeyClaude submission-field factory. Use it for source and provenance review when a content, package, or registry intake needs evidence about where a submission came from, how an artifact was produced, and whether trust claims match the expected source.
Core Workflow
- Classify the submission as source-only content, package metadata, generated output, or a release artifact.
- Inventory evidence before judgment: canonical docs, repository, commit, tag, release, package metadata, artifact digest, provenance record, verification summary, and submitter claims.
- Establish expectations from policy: trusted producer, repository, source revision, branch or tag, artifact name, digest algorithm, builder ID, build type, external parameters, and required SLSA level.
- Review artifact provenance when an artifact is present: subject digest, provenance signature status, trusted builder ID, expected build type, expected source repository, expected source revision, and allowed external parameters.
- Review source provenance when the claim is about a repository revision: source control system, revision immutability, change history continuity, technical controls, code review claims, and verification summaries if available.
- Compare trust claims to evidence. Separate confirmed facts, unsupported claims, stale evidence, unverifiable links, and policy gaps.
- Decide the intake path: accept, accept with caveats, request more evidence, route to manual review, or reject as unverifiable or mismatched.
- Produce a compact provenance report with source URLs, checked identifiers, failed expectations, residual risks, and the final recommendation.
Capability Scope
- Source repository and revision provenance review
- Build provenance and artifact evidence checks
- Package or generated-output source review
- SLSA expectation mapping for intake policy
- Digest, builder, predicate, and source-URI comparison
- Evidence classification for maintainer review
- Unverifiable, stale, or overclaimed trust-signal detection
- Public-safe verification notes for PRs, registry entries, and package listings
Compatibility
Native
- Claude Code / Claude: use as a reusable Agent Skill for provenance-heavy submission review.
- Codex/OpenAI workflows: use as
SKILL.md-style instructions for registry, package, or artifact intake.
Manual Adaptation
- Windsurf and Gemini: adapt the workflow and output contract into their skill formats.
- Cursor and Generic AGENTS files: convert the production rules and validation checklist into repository-level intake rules.
Required Inputs
- Submission category and acceptance policy
- Canonical source URL, repository URL, docs URL, release URL, or package URL
- Commit SHA, tag, branch, release version, package version, or artifact digest
- Published provenance, verification summary, or package ecosystem trust metadata when available
- Expected producer and builder IDs
- Decision threshold for missing or partial provenance
Production Rules
- Do not treat provenance as a replacement for source review, license review, or category-fit review.
- Do not accept a package-trust claim when the artifact digest does not match the provenance subject.
- Do not accept a source-trust claim when the source repository, source revision, or producer is ambiguous.
- Do not infer SLSA level from marketing copy. Require published evidence or mark the claim unsupported.
- Prefer immutable identifiers such as commit SHAs, release tags with tag object evidence, artifact digests, and package integrity metadata.
- Treat branch names, moving tags, latest URLs, and generated archive links as weaker evidence unless backed by stronger provenance.
- Keep public notes focused on non-sensitive identifiers and high-level risk. Avoid publishing raw logs or operational context that is not needed for review.
Decision Rubric
Accept
- Canonical source and artifact identifiers are clear.
- Digest, repository, revision, builder, and predicate expectations match.
- Published provenance or source verification evidence is reachable and current.
- Safety, privacy, and trust notes reflect the verified evidence.
Request More Evidence
- The source is real but the artifact digest, builder ID, release tag, or verification summary is missing.
- Provenance exists but cannot be linked to the submitted artifact or source revision.
- The submission makes trust claims that require maintainer confirmation.
Reject Or Route Away
- The source URL is unverifiable, redirects to unrelated content, or conflicts with package metadata.
- The provenance subject does not match the artifact under review.
- The builder, source repository, or revision does not match the expected producer.
- The submission asks maintainers to validate a generated artifact without source-backed trust evidence.
Output Contract
- Evidence inventory: source URL, repo, revision, package, artifact, digest, provenance, and verification-summary links.
- Expectations: producer, source repository, revision, builder, predicate, SLSA level, and policy threshold.
- Review result: matched evidence, failed checks, missing data, stale links, and unsupported claims.
- Safety and privacy notes: runtime risk, package or generated-output risk, public-note redactions, and residual uncertainty.
- Intake decision: accept, accept with caveats, request evidence, manual review, or reject.
- Follow-up: exact evidence needed from the submitter or maintainer before the decision can change.
Workflow Notes
- Keep automated provenance checks read-only in public PR workflows.
- Use maintainer-controlled workflows for any privileged package, generated output, or provenance review.
- Store review results as summaries, not raw operational logs.
- Re-run review when the artifact, release tag, package version, builder, or source revision changes.
Troubleshooting
Issue: The artifact has no provenance
Fix: Record the missing evidence and fall back to source, package, release, and maintainer judgment. Missing provenance may be acceptable for low-risk content but should block strong verification claims.
Issue: The provenance exists but the digest does not match
Fix: Treat the artifact as unverified. Ask for the exact artifact referenced by the provenance record or a new provenance record for the submitted artifact.
Issue: The source revision is a branch name
Fix: Resolve the branch to an immutable commit SHA, record when it was checked, and prefer a tag or release object when reviewing a published package.
Issue: The submitter claims SLSA compliance without evidence
Fix: Mark the claim unsupported. Require reachable source, builder, predicate, and verification evidence before using the claim in public metadata.
Issue: The evidence includes details that do not need to be public
Fix: Summarize only the identifiers needed for review and keep unnecessary operational details out of public issues or PRs.
Validation Checklist
- Source, repo, package, release, and artifact URLs are reachable.
- Artifact digest and provenance subject match when an artifact is reviewed.
- Builder ID is one of the trusted builders for the expected policy.
- Build type and external parameters match expected values.
- Source repository and source revision match the submitted package or artifact.
- Source provenance or verification summary evidence is recorded when source-level claims are made.
- Missing or partial provenance is called out as a caveat, not hidden.
- Public notes do not expose operational details that are unnecessary for review.
Source citations
Signals
Loading live community signals…
A short, calm digest of reviewed Claude resources. Unsubscribe any time.