The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's 'Resume' mode establishes a pattern where the agent reads instructions from a local markdown file and is directed to treat 'Locked decisions' as settled and 'Invoke suggested skills' based on the file content.
Ingestion points: The skill reads markdown files from .handoff/ or docs/handoffs/ during the resume workflow.
Boundary markers: The skill uses specific markdown headers (e.g., '## Locked decisions', '## Suggested skills') to structure the imported context.
Capability inventory: The agent can write files, modify .gitignore via shell commands, and invoke other installed skills as directed by the handoff document.
Sanitization: The skill includes explicit instructions for the agent to redact secrets (API keys, PII, tokens) before writing the handoff and to verify pointers and assumptions when resuming.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with the local environment to ensure session privacy.
Evidence: In SKILL.md, the workflow instructs the agent to check and update the .gitignore file to include the .handoff/ directory if it is not already present.

session-handoff