The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify external skill files and evaluation configurations, creating an attack surface where malicious instructions embedded in those files could influence the agent.
Ingestion points: The agent reads SKILL.md, .claude/upstream.md, and evaluations/<skill-name>.json (Steps 1 and 6).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands within the ingested files are specified.
Capability inventory: The agent has the authority to execute repository commands, including npm run validate, git log, skill-validator, and node scripts/evaluate-skills.js (Steps 1, 2, 5, and 7).
Sanitization: The instructions do not describe sanitization or filtering of the content read from the files before processing.
[COMMAND_EXECUTION]: The skill instructs the agent to perform several shell-based operations and execute local scripts to maintain project quality and verify changes.
Operations: Execution of npm run validate for project consistency, git log for change history, skill-validator for LLM-based quality scoring, and node scripts/evaluate-skills.js for running regression tests.

improve-ic-skill