SkillSpector: documented skill-security analysis and triage
SkillSpector documents static and optional LLM-assisted analysis of agent-skill contents, followed by risk reports and user-defined triage.
L2Research workflow (workflow level, not a quality rating) Source: Repository-documented Outcome: Not empirically evaluated here
Research applicability
A documented static-plus-LLM skill-security analysis and triage reference for agent evaluation.
Getting started
Read the linked primary sources and the stated adoption requirements. Reuse suggestions are editorial and untested; no setup recipe or trial was executed.
Maintenance
Sources last checked on 2026-10-19. Future availability, maintenance, provider terms and compatibility were not verified.
License
The queued README states Apache 2.0; full licence, dependency and reuse terms were not audited.
Workflow & human oversight
Intended documented L2 for one SkillSpector reference: users select inputs and mode; static and optional semantic stages produce reports; users accept known findings and set gate policy. Capabilities are bound to the original GitLab README snapshot. Related copies are not separate products, and current upstream differences are not merged into that snapshot.
Workflow level is not an evidence score; higher is not better evidence.
Approach
The selected queued documentation connects static pattern/AST checks, dependency lookup and optional semantic assessment to structured findings and user-controlled baselines/install policy. This is one resource reference across known copy identities. Its snapshot says semantic analysis filters false positives and recognized signature bundles are excluded; current upstream instead documents retaining deterministic findings and analysing DSSE contents. These behaviours are version-specific source claims, not merged guarantees.
Limitations
Queued GitLab README/metadata and selected current upstream pipeline/trust/limitation passages only. Moving HEADs are unpinned and non-atomic. The queued snapshot says semantic analysis filters false positives and recognised signature bundles are excluded; upstream instead describes retaining deterministic findings and analysing DSSE contents. No official GitLab affiliation, distinct fork contribution or current implementation certification is inferred. Earlier related-copy matches do not prove present equivalence. Headline dataset vulnerability rates and precision estimates were not validated. LLM mode sends file contents to its configured provider; static-only mode can still send dependency coordinates to OSV. No sandbox, signature validity, complete coverage, safety or runtime assurance is established. No scanner, model or provider was run.
Reproducibility
Source text was checked only within this scope: Original queued README and identity metadata, plus selected differing upstream documentation; no implementation audit. No software, dataset, experiment or result was independently reproduced.
Adoption requirements
Untested suggestion: select and pin a source revision, confirm finding-retention/signature/completeness behaviour, check egress and data rights, and validate reports against known findings before a research trial.
Comments
No comments yet.
Add a comment
Comments are held for moderation before appearing. Critical or corrective comments are welcome — especially evidence that an entry overstates what has been verified.