The Agent Skills Directory

[SAFE]: The skill provides comprehensive guidelines for prose editing and local data verification. The included Python script scripts/validate_receipt.py is a utility for checking data consistency and does not perform network operations or execute untrusted code.
[SAFE]: The validation script follows secure coding practices by implementing path traversal mitigations (checking for absolute paths, '..' components, and ensuring relative paths stay within the root directory). This prevents the script from being used to read sensitive system files outside of the designated receipt folder.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text for rewriting, which is an inherent attack surface for indirect prompt injection.
Ingestion points: Original prose text and supplied facts are ingested via the skill's primary instructions in SKILL.md.
Boundary markers: Although formal delimiters are not defined in the prompt template, the workflow requires the agent to inventory and verify all claims against trusted sources before processing, acting as a functional boundary.
Capability inventory: The agent's capabilities are restricted to text transformation and local file hashing via the provided script; it lacks permissions for network data transmission or arbitrary command execution.
Sanitization: Instructions in references/measurement.md explicitly direct the user to remove sensitive credentials, session URLs, and account details before artifacts are processed or shared.

anti-detection-writing