The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data including git diffs, pull request metadata, and meeting transcripts to generate its documentation and articles.
Ingestion points: As described in SKILL.md and reference/article-handoffs.md, data enters the context via the 'Launch' tool (PR info), 'Trail' tool (git history), and direct user uploads of transcripts or drafts.
Boundary markers: The skill uses structured handoff templates with YAML delimiters and mandates the use of [Inference: evidence] markers to separate inferred intent from technical facts found in the code.
Capability inventory: The skill can execute git commands (e.g., git reflog) and write various documentation files.
Sanitization: The 'Boundaries' section in SKILL.md explicitly forbids including security implementation details (secret keys, auth internals) and requires masking internal project details before public release.
[COMMAND_EXECUTION]: The skill's documentation and reference files mention several command-line tools and shell operations.
Evidence: References to git reflog, npx zenn-cli, brew install, and npx zenn-cli init appear in SKILL.md and reference/article-platform-optimization.md.
Context: These commands are contextually appropriate for the developer-centric purpose of the skill. The sudo mention in reference/quickstart-guide.md is part of a troubleshooting template designed for human users to follow, rather than an instruction for the agent to escalate its own privileges.

tome