The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved from the StoryLoop platform, creating a surface for potential instructions embedded in story content to influence the agent.
Ingestion points: Data enters the context via inspect_story, inspect_story_history, and get_planning_session as seen in SKILL.md and references/planning.md.
Boundary markers: SKILL.md instructs the agent to keep questions and assumptions distinct from requirements and to treat tool state as authoritative over transcripts.
Capability inventory: Capabilities include performing MCP tool mutations and managing implementation via the host harness and GitHub repositories.
Sanitization: The skill enforces strict output filtering by forbidding the transmission of shell commands, local file paths, secrets, or deployment credentials to the StoryLoop platform.

storyloop