The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to discover and execute shell commands from the project environment to establish baselines and validate changes.
Evidence: In SKILL.md, Step 2 requires the agent to "Run the relevant baseline commands before editing, using the same environment and selection intended for final validation."
Evidence: In SKILL.md, Step 5 requires running the "smallest affected suite, then relevant sibling and cross-package consumers and repository-required gates. Use commands discovered from the project."
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (project source code, test files, and engineering documentation) and uses this data to inform its decision-making for file modifications and command execution.
Ingestion points: The agent reads manifests, runner configurations, CI definitions, engineering docs, and production source code (SKILL.md, Step 2).
Boundary markers: The instructions do not specify the use of delimiters or safety warnings when processing these external files.
Capability inventory: The skill has the authority to edit test files (SKILL.md, Step 4) and execute arbitrary shell commands discovered in the project (SKILL.md, Step 2 and 5).
Sanitization: No sanitization or validation of the ingested content is described before it influences agent actions.
[AUTONOMY]: The instructions encourage the agent to perform multiple edits without per-change confirmation once a high-level "cleanup" task is started.
Evidence: SKILL.md states in Step 1: "Inspect first, then continue without asking for approval already given." while Step 4 and 5 involve applying deletions and repairs based on its own earlier analysis.

test-audit