The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses behavioral constraints to override default agent interaction patterns, specifically instructing the agent to ask "no questions" and use "no narration" to minimize delay.
[COMMAND_EXECUTION]: The skill invokes an external tool named session_capability_inventory during the extended sync process to regenerate project inventory data.
[DATA_EXFILTRATION]: The skill possesses an indirect prompt injection surface by reading and processing local file content without explicit sanitization or boundary markers.
Ingestion points: Reads CLAUDE-capability-inventory.md, MEMORY.md, and various session logs.
Boundary markers: None present; the skill treats file content as direct input for its summary generation.
Capability inventory: The skill has the capability to write to multiple files in the session-context/ directory and MEMORY.md.
Sanitization: No sanitization or validation of the ingested file content is performed before interpolation into the output summary.

sync