The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read data from external sources such as Pull Requests, commit messages, and Issue tickets to understand code intent. These sources are considered untrusted as they could contain instructions designed to manipulate the agent.
Ingestion points: PRs, Issues, and commit messages (as described in step 2 of SKILL.md).
Boundary markers: No explicit markers or delimiters are provided to wrap or isolate the untrusted external content.
Capability inventory: The agent has the capability to write to the filesystem (git stage/commit) and execute local commands (automated checks).
Sanitization: No sanitization or validation of the ingested intent text is performed.
[DYNAMIC_EXECUTION]: The skill directs the agent to identify and run automated project checks such as tests, type-checkers, and formatters. This involves executing project-defined scripts and tools locally.
Execution pattern: Running typecheck, tests, and format commands discovered within the project environment.
Risk assessment: This is standard behavior for development-focused skills and does not involve remote code execution from unknown sources.

resolving-merge-conflicts