The Agent Skills Directory

[SAFE]: The skill implements robust path containment in scripts/inventory.mjs by validating that all file paths, when resolved to their real paths, remain within the repository's top-level directory.
[SAFE]: Git commands and script invocations are executed securely; user-supplied pathspecs are passed to git using the :(literal) magic prefix and the -- separator to prevent pathspec and flag injection.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted code and coverage data. 1. Ingestion points: scripts/inventory.mjs reads tracked files and lane analysts process source code from the repository. 2. Boundary markers: references/lane-analyst.md requires the lane analyst to return data in a specific fenced JSON block to maintain structure. 3. Capability inventory: The skill is limited to read-only file access and specific git commands; lane analysts are explicitly restricted to read-only fresh contexts. 4. Sanitization: shared/external-data.md provides comprehensive rules for handling untrusted prose, including explicit instructions for the agent to report embedded imperatives without acting on them.

auditing-complexity