tool-call-configuration-for
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.runinscripts/lib/apply_plan.pyto execute validation commands provided in an integration plan. This behavior is restricted to the defined target directory and is used for running project-specific CI tasks like lints and tests after an update. - [DATA_EXPOSURE]: The skill accesses local files such as tool descriptors, behavior rules, and existing skill directories. These operations are protected by a
confinedhelper function that prevents path traversal and ensures the agent cannot read or write files outside the declared skills root. - [INDIRECT_PROMPT_INJECTION]: The skill ingests external behavior instructions (markdown, YAML, or JSON) and interpolates them into newly generated
SKILL.mdfiles. While this creates an attack surface for instructions in generated skills, the generator includes explicit boundary markers and documentation stating these are instruction-only rules without host-level enforcement. - [SAFE]: Integrity is maintained through SHA-256 hash checks. The
applycommand refuses to modify a target if its current state does not match the expected hash in the integration plan, protecting against concurrent modification conflicts.
Audit Metadata