The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process information from active agent sessions to update a persistent user profile.
Ingestion points: Data enters the system via the update-from-session command in SKILL.md and user-provided form submissions in scripts/onboarding_webui.py.
Boundary markers: The skill instructions and the references/update-policy.md documentation mandate a strict separation between durable profile facts and temporary task-local instructions (e.g., repository rules or system prompts).
Capability inventory: The skill possesses the capability to write to a local SQLite database and host a temporary local web server for configuration.
Sanitization: The skill employs a comprehensive SECRET_PATTERNS regex suite to automatically detect and redact credentials (tokens, private keys, passwords) before storage. Furthermore, inferred or sensitive data is placed in a 'pending' state, requiring explicit user confirmation via the proposal-apply tool before becoming active.
[COMMAND_EXECUTION]: The skill uses local Python scripts to facilitate its operations.
Evidence: SKILL.md and references/examples.md demonstrate the use of scripts/profile_store.py for all database management tasks and scripts/onboarding_webui.py for the interactive onboarding process.
[DATA_EXPOSURE]: The skill implements strong local data protections.
Evidence: As defined in scripts/profile_store.py and references/privacy-boundary.md, the skill enforces POSIX file permissions (0700 for directories and 0600 for the SQLite database) and restricts data access to specific 'views' to prevent the accidental exposure of private information to other agent skills.

user-profile-keeper