The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to clone external repositories and execute their local test suites and linters during the verification gates (referenced in SKILL.md Gate 10 and Gate 12). This involves executing code from arbitrary, potentially untrusted repositories.
[DYNAMIC_EXECUTION]: The skill invokes local tools and scripts, such as a "queue's poster" and "hash helper," which are expected to reside in a separate local clone of a private repository rather than within the skill's own directory (referenced in SKILL.md Step 11).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external sources, which could influence the agent's operations.
Ingestion points: The agent retrieves issue bodies, comments, and pull request descriptions from arbitrary repositories via the GitHub API (referenced in SKILL.md Gates 5, 6, and 8).
Boundary markers: No specific boundary markers or instructions to treat external data as untrusted are provided in the prompt logic.
Capability inventory: The agent has access to the gh CLI for network operations, git for filesystem and network operations, and the ability to execute local commands for testing.
Sanitization: The instructions do not specify any sanitization or validation of the content retrieved from external repository threads.

oss-contribution-research