The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically identifies and executes local build and test commands (such as npm test, pytest, cargo test, or make targets) based on configuration files found in the repository. These commands are executed via subprocess.run to verify code changes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external GitHub PR Checks and Action runs, which could potentially contain malicious instructions embedded by external contributors in check names or summaries.
Ingestion points: fetch_failing_checks.py retrieves PR check details and check-run summaries from the GitHub API using the gh CLI.
Boundary markers: Data retrieved from GitHub is formatted into a markdown report (pr-<N>-ci-fails.md) without explicit instruction delimiters or "ignore embedded instructions" warnings.
Capability inventory: The skill is designed to automatically modify repository files ("apply fixes") and execute shell commands discovered in configuration files via discover_and_verify.py.
Sanitization: There is no evidence of sanitization or filtering of the state, description, or summary fields fetched from the GitHub API before they are written to the report and processed by the agent.

heyeddi-ci-fails