The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to distill and persist reusable lessons into agent instruction files, which is a mechanism that could be targeted to influence future agent behavior.
Ingestion points: Processes user requests and existing project documentation to extract and format new rules.
Boundary markers: The execution boundary in SKILL.md explicitly separates analysis and preview from modification, requiring user confirmation for persistence.
Capability inventory: Modifies instruction files and writes to a local profile JSON using scripts/profile_store.py.
Sanitization: scripts/profile_store.py includes a filter (SENSITIVE_PARTS) to prevent the persistence of secret-like fields (tokens, passwords, keys) into the long-term profile storage.
[COMMAND_EXECUTION]: The skill includes and utilizes local Python scripts for operational tasks.
Evidence: scripts/profile_store.py is used to manage identity and brand configuration data in the local filesystem.
Evidence: scripts/validate_skill.py performs structure and metadata linting of the skill source code using PyYAML.
[SAFE]: The external references and dependencies are standard for this ecosystem and follow secure management practices.
Evidence: skill.yaml defines a clear user-profile/v1 schema for data persistence with a restricted namespace.

lov-distill-to-system