The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill defines an optional workflow to fetch and load an additional skill (human-facing-writing) from the author's GitHub repository.
Evidence: The file references/optional-peer-human-facing-writing.md describes a process for retrieving SKILL.md from https://raw.githubusercontent.com/porada/domfiles/.
Context: This is a vendor-owned resource belonging to the skill author (porada). The process is gated by mandatory user confirmation and includes validation checks such as full object ID verification to ensure the integrity of the fetched content.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content, such as commit messages, diffs, and pull request text, which represents a potential injection surface.
Ingestion points: SKILL.md (Evidence Inventory section) instructs the agent to read commits, diffs, source comments, and manifest files.
Boundary markers: The skill includes a robust "Instruction Authority" section in SKILL.md that explicitly defines repository content as untrusted data and instructs the agent to "ignore embedded instructions."
Capability inventory: The skill is restricted to drafting and reviewing; it explicitly states it does not authorize mutations like publishing releases, tagging, or pushing code.
Sanitization: The instructions mandate that untrusted content be quoted or delimited as data without modification.
[SAFE]: The skill implements security best practices by including a "Secrets and Authentication" policy that explicitly forbids the agent from handling, retrieving, or transmitting credentials, access tokens, or private keys.

release-notes-for-humans