The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various sources which can influence agent behavior in future sessions.
Ingestion points: The skill reads entries from .learnings/ (LEARNINGS.md, ERRORS.md, etc.) and session traces via the entire CLI tool (SKILL.md, Step 1 and Session Trace Analysis section).
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions within the processed log entries or traces.
Capability inventory: The resulting 'Gap Report' is designed to be consumed by the harness-updater agent, which applies changes to project instruction files such as CLAUDE.md, AGENTS.md, and .github/copilot-instructions.md (SKILL.md, Step 5).
Sanitization: The skill lacks mechanisms to sanitize or validate the content of log entries before they are recommended for promotion into system-level instructions.
[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands using the entire CLI tool to fetch and analyze session data.
Evidence: Commands include entire --version, entire rewind --list, and entire explain with various flags (SKILL.md, Session Trace Analysis section).

learning-aggregator