The Agent Skills Directory

[PROMPT_INJECTION]: The skill scans project documentation and session history to generate recaps and memory updates, which creates a surface for indirect prompt injection.
Ingestion points: Processes CLAUDE.md, README.md, MEMORY.md, and session history content.
Boundary markers: The protocol does not specify the use of delimiters or 'ignore embedded instructions' warnings when summarizing these files.
Capability inventory: Includes the ability to write to multiple file paths, call other skills, and perform git push operations.
Sanitization: No explicit sanitization or filtering of external file content is described before processing.
[COMMAND_EXECUTION]: The skill uses local git commands (status, diff, rev-parse, remote) to inspect repository state. Actions that modify the repository (commit and push) are user-confirmed through multi-select prompts.

session-recap