The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill triages and edits bug reports, which constitute an untrusted data ingestion surface.
Ingestion points: The bug report file is loaded in Step 1 of SKILL.md.
Boundary markers: The subagent prompts in references/review-bug-prepass-prompts.md strictly limit the agents to read-only review and YAML output.
Capability inventory: The skill can execute git commands, Node.js/Python scripts, and perform network operations via established tracker helpers.
Sanitization: The skill includes logic to redact hardcoded credentials and sanitize control characters in issue bodies.
[COMMAND_EXECUTION]: The skill uses local CLI tools including git, node, python, and gh to manage branches, read configurations, and post tracker comments. Commands are invoked using safe implementation patterns (such as passing arguments as arrays) to prevent shell injection.
[SAFE]: The skill demonstrates high security standards, incorporating internal defenses against common attack vectors such as local Python environment shadowing and accidental credential exposure in logs.

review-bug