The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources including chat history, shared records (via the why skill), and GitHub issues/PRs via gh.
Ingestion points: SKILL.md (Steps 3, 4, 5) and recall-procedure.md (Steps 3, 4, 5) describe reading chat transcripts, issue trackers, and documentation to reconstruct context.
Boundary markers: The skill lacks instructions for using robust delimiters (such as XML tags or unique markers) when interpolating external data into prompts, which could allow data to be interpreted as instructions.
Capability inventory: The skill utilizes git and gh tools and is authorized to spawn subagents for parallel processing.
Sanitization: The instructions include a high-level directive to 'Sanitize private context before any public output.'
[COMMAND_EXECUTION]: The skill utilizes system-level commands to gather project context.
Evidence: SKILL.md (Step 5) and recall-procedure.md (Step 5) instruct the agent to use git and gh (GitHub CLI) to verify the live state of the project. This is consistent with the skill's documented purpose for recalling work status.

recall