The Agent Skills Directory

[SAFE]: The skill performs legitimate file operations within the workspace, such as reading resume.yaml and writing tailored cover letters to a local knowledge/ directory. No network activity, credential access, or administrative commands were detected.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data in the form of Job Descriptions (JDs) provided by the user. While this creates a surface for indirect prompt injection (e.g., instructions hidden in a JD), the risk is mitigated by explicit "Anti-Fabrication Rules" and a mandatory "Claim Verification" step.
Ingestion points: User-provided Job Description text (Step 1).
Boundary markers: None explicitly defined for the input text, but processing is constrained by strict factual rules.
Capability inventory: Reading and writing files in the project workspace; loading local agent instruction files (agents/).
Sanitization: Step 5 enforces a manual verification loop where every claim must be traced back to resume.yaml, effectively filtering out instructions or fabrications injected via the JD.

cover-letter