The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from Pull Request titles, bodies, commit messages, and comments, as well as external Linear issue data to extract identifiers.
Ingestion points: PR metadata (GitHub), commit messages (Git), and issue details (Linear MCP).
Boundary markers: The skill does not explicitly use delimiters to separate instructions from ingested data; however, it uses narrow regex patterns to extract specific formats (e.g., Linear IDs or Slack URLs).
Capability inventory: The agent has write access to the GitHub repository via the GitHub MCP to post comments and manage labels. It also uses the git tool to read memento notes.
Sanitization: Data is processed via regular expressions to extract specific identifiers before being formatted into a JSON manifest, which limits the risk of raw content influencing the agent's logic.
[COMMAND_EXECUTION]: The skill uses standard Git commands to check for memento notes on commits.
Evidence: Uses git fetch and git notes show {sha} in Steps 5 and 9 to manage knowledge tracking metadata stored in Git notes.

mark-evidence