The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull requests, issues, and discussions. This content, which can be provided by any external contributor, is passed to subagents for analysis. While the skill uses structural boundaries by requiring JSON-only output from subagents, it lacks explicit sanitization or filtering of the ingested content. \n
Ingestion points: Data is fetched via gh pr view, gh issue view, and gh api graphql from repository content.\n
Boundary markers: Subagent prompts use strict formatting instructions: 'Return ONLY a JSON object on one line (no prose, no fences)'.\n
Capability inventory: The skill executes gh CLI commands and performs file system writes to generate reports and snapshots.\n
Sanitization: No sanitization of GitHub item bodies or comments is performed before processing.- [DATA_EXFILTRATION]: The skill includes an opt-in feature to publish triage reports to a GitHub Gist using gh gist create. Although this action is user-confirmed and targets a well-known service, it involves exporting repository metadata and contributor information to an external endpoint.- [COMMAND_EXECUTION]: The skill relies extensively on the gh CLI to interact with GitHub's REST and GraphQL APIs, as well as to manage local file state (e.g., updating .gitignore).

triage-github