The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-supplied content or existing text to generate social media copy, which creates an attack surface for instructions embedded within that data.\n
Ingestion points: SKILL.md workflow step 1 reads user-specified content or currently confirmed text.\n
Boundary markers: The instructions in SKILL.md and references/authorship-integrity.md specify that the agent should only write supported experiences, maintain accurate brand names, and preserve specific verbatim text.\n
Capability inventory: The skill possesses filesystem write capabilities (SKILL.md step 4) to save output files and read/write capabilities via scripts/profile_store.py.\n
Sanitization: The workflow requires avoiding markdown formatting noise and explicitly forbids fabricating links or personal information not supported by the source.\n- [PERSISTENCE]: The skill implements a persistent profile management system using the scripts/profile_store.py script. This script writes user-stated preferences and identity facts to a local JSON file (e.g., ~/.config/agent-skills/profile.json), allowing state to be maintained across sessions.\n
Evidence: The profile_store.py script uses atomic_write to replace the profile file and set_path to update specific fields. It includes a security check to reject paths containing sensitive keywords like 'token', 'secret', or 'password'.

lov-output-wechat-moment