seb1n/awesome-ai-agent-skills / agent-security/skill-supply-chain-audit
skill-supply-chain-audit
VerifiedAudit agent skills, plugins, prompts, manifests, scripts, dependencies, and bundled assets for provenance, prompt-injection, permission, execution, exfiltration, persistence, and update risk. Use when evaluating a third-party skill before installing, enabling, updating, publishing, or distributing it; reviewing an untrusted SKILL.md, agent configuration, MCP integration, archive, or repository; comparing a package with a known-good version; or investigating unexpected tool, network, credential, or filesystem behavior.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.6 KB |
| agents/openai.yaml | config | 223 B |
| assets/audit-report-template.md | markdown | 1.7 KB |
| references/review-checklist.md | markdown | 3.6 KB |
| scripts/audit_skill.pyexecutable | script | 34.8 KB |
references/review-checklist.md
View on GitHub# Static review checklist
Load this reference for a full audit or when assigning severity.
## Provenance and integrity
- Resolve the exact source, version, commit, publisher claim, release date, and license.
- Hash every regular file and record the canonical package-manifest hash; note missing, unverifiable, or changing artifacts.
- Identify generated, vendored, minified, encoded, encrypted, binary, or archived content.
- Inspect ZIP/TAR member metadata without extraction. Check member names for absolute paths, drive paths, parent traversal, links, and special entries; bound member count, member size, declared total expansion, and compression ratio.
- Resolve every file in the scanner's `content_review_queue` using a bounded read-only chunked/manual review, or retain it as an explicit disposition-blocking coverage gap.
- Compare the release with a known-good baseline and upstream source when available.
## Instruction integrity
- Does metadata describe the actual behavior and activation scope?
- Can retrieved content, filenames, comments, issue text, or documents become instructions?
- Do instructions request secrecy, authority bypass, false completion, or skipped verification?
- Are approval gates preserved for external writes and high-impact actions?
- Are users told when information is inferred, unverified, stale, or incomplete?
## Code and installation
- Enumerate entry points, executable bits, interpreters, lifecycle hooks, and subprocesses.
- Review dynamic execution, deserialization, templated shell commands, and unsafe temporary files.
- Review deletes, overwrites, permission changes, persistence, and writes outside the target.
- Review network domains, redirects, uploads, telemetry, webhooks, and remote installers.
- Review credential stores, environment variables, browser/session data, and cloud metadata access.
- Review manifests, lockfiles, dependency sources, version ranges, hashes, and name confusion.
## Tools and MCP
- Match each declared tool to a necessary user-visible capability.
- Verify server ownership, transport security, authentication scope, and data retention.
- Check whether tool output is treated as untrusted data.
- Require confirmation for messages, deployments, purchases, account changes, and destructive writes.
## Data and privacy
- Map collected fields, purpose, destination, retention, and deletion path.
- Minimize access to personal, confidential, regulated, or customer data.
- Distinguish local processing from upload or third-party processing.
- Check logs, caches, crash reports, and generated artifacts for secret leakage.
## Severity
- **Critical:** credible path to arbitrary execution, credential theft, destructive action, covert persistence, or material exfiltration without a meaningful prerequisite.
- **High:** major confidentiality, integrity, authority, or availability impact with realistic prerequisites.
- **Medium:** bounded impact, meaningful hardening gap, or permission expansion requiring multiple conditions.
- **Low:** limited weakness, clarity issue, or defense-in-depth gap.
- **Informational:** inventory fact or improvement with no demonstrated security impact.
Severity describes impact and likelihood; confidence describes evidence quality. Do not combine them. Separately label the claim itself `observed`, `inferred`, or `unknown`.
## Minimum finding record
For each finding include: ID, title, severity, confidence, claim status (`observed`, `inferred`, or `unknown`), remediation status, evidence with file and line or hash, behavior, exploit prerequisites, impact, affected versions, remediation, verification step, and residual risk.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |