The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests case information (titles, descriptions, and metadata) from external repositories via the gh CLI. If a repository's content is controlled by an attacker and contains hidden instructions, the agent might process them during the browsing workflow.
Ingestion points: External repository cases directory metadata fetched via gh CLI.
Boundary markers: The SKILL.md mentions that analysis and preview requests should remain read-only, but there are no specific prompt delimiters defined for handling external case data.
Capability inventory: The skill has capabilities for local file writing (via profile_store.py) and CLI execution (via gh).
Sanitization: The profile_store.py script includes a blacklist (SENSITIVE_PARTS) to prevent writing keys like api_key, secret, or password into the durable profile.
[COMMAND_EXECUTION]: The skill requires and invokes the gh (GitHub) command-line interface to retrieve case lists and verify repository status. This is a standard dependency for the skill's stated purpose of browsing repositories.
[PERSISTENCE]: The included script scripts/profile_store.py is designed to write and update a local JSON file located at ~/.config/agent-skills/profile.json (or similar paths). This is used to maintain 'durable records' of user preferences and identity across sessions.

lov-xbti-gallery