The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external Validation Reports which could contain malicious instructions. It mitigates this risk by explicitly instructing the agent to treat findings as claims rather than instructions and to verify all claims against the actual code before taking action. Mandatory Evidence Chain:\n * Ingestion points: Reads reports from the conversation context or prior-artifact files in SKILL.md.\n * Boundary markers: Instructions to restate and classify findings provide logical separation.\n * Capability inventory: Uses Bash, Edit, and Write tools for code modification and verification.\n * Sanitization: Rule 1 in the reference file explicitly labels external input as untrusted and Rule 3 requires reproduction of behavior claims before implementation.\n- [COMMAND_EXECUTION]: The skill executes shell commands (e.g., bun run check) to validate fixes. This is a legitimate use of the tool within a developer environment for its stated purpose.\n- [SAFE]: The skill follows security best practices by requiring reproduction of issues before applying fixes and limiting the scope of changes to the smallest possible diff. No malicious patterns detected.

remediate-plan