The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local preparation packets and body files, which could theoretically contain malicious instructions if the local files are compromised. The skill mitigates this through a mandatory review process.
Ingestion points: Reads local files containing prepared contribution packets and body content via the Read tool.
Boundary markers: Implements a 'Mandatory approval boundary' that requires the agent to display the full repository target, command, content, commit SHA, and test evidence before execution.
Capability inventory: Capable of creating issues, PRs, comments, and pushing branches using Bash(gh:*) and Bash(git:*).
Sanitization: Relies on explicit human verification of the displayed content and target as a final safety gate before any mutation is performed.
[COMMAND_EXECUTION]: The skill utilizes system-level commands through gh (GitHub CLI) and git to perform its primary function. It operates within the scope of the user's existing authentication and includes instructions to never request, read, or print sensitive tokens, minimizing the risk of credential exposure during command execution.

contribute-publish