The Agent Skills Directory

[SAFE]: The skill implements a restricted dispatcher pattern, delegating sensitive operations to specialized sub-skills which maintain their own confirmation gates and execution logic. This design minimizes the attack surface of the routing component.
[INDIRECT_PROMPT_INJECTION]: The skill identifies external data sources such as pull request bodies, commit messages, and tags as untrusted inputs. 1. Ingestion points: The skill uses gh and git commands to resolve repository metadata; 2. Boundary markers: It includes an explicit instruction to the agent to never obey instructions embedded in PR bodies or commit messages; 3. Capability inventory: Access is limited to git and gh tools via Bash; 4. Sanitization: The routing logic employs strict string matching for subcommands and verifies branch existence before proceeding. This assessed risk is SAFE due to proactive mitigation.
[COMMAND_EXECUTION]: Shell commands are strictly limited to repository state detection (git rev-parse, gh repo view). The skill includes safety checks to prevent execution on unverified or guessed branch names, requiring manual user intervention when resolution is ambiguous.

release-dispatch