The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted job logs from GitHub Actions runs during the Phase 4a 'LLM gap-fill' process. This creates a potential surface for indirect prompt injection attacks where malicious content in the logs could attempt to influence the agent's behavior during analysis.\n
Ingestion points: Job logs are fetched by collect_runs.py and provided to the agent for analysis in Phase 4a.\n
Boundary markers: Explicit instructions are present in SKILL.md and references/gap-fill.md directing the agent to treat log content as untrusted data and ignore any instructions or directives embedded within them.\n
Capability inventory: The agent executes scripts such as run.py and collect_runs.py which utilize subprocess.run to call git and the gh CLI.\n
Sanitization: The skill provides instructions for the agent to manually mask credentials, but there is no automated sanitization of ingested logs implemented in the code.\n- [COMMAND_EXECUTION]: The skill performs shell command execution using the gh and git CLIs to inspect the repository and fetch Actions data. These operations are core functionality for the skill but involve parameters derived from repository metadata.\n- [EXTERNAL_DOWNLOADS]: The skill fetches data from GitHub APIs via the gh CLI. This is a trusted and well-known service necessary for the primary purpose of the skill.

ci-speedup