The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands using the GitHub CLI (gh api). These commands are used for telemetry management, specifically for creating, updating, and reconciling status comments on repository issues. The implementation includes logic to handle race conditions and synchronization without involving unsafe user-controlled command interpolation.
[EXTERNAL_DOWNLOADS]: The skill accesses local configuration files, such as ~/.claude/rate-limit-guard/rate-limits.json, to monitor API usage limits. It also references internal project documentation and sibling plugin resources within the ${CLAUDE_PLUGIN_ROOT} directory to maintain shared context for the triage lanes.
[PROMPT_INJECTION]: The skill manages potential indirect prompt injection through several mechanisms. Its ingestion points include reading issue titles, bodies, and comments from the target repository. It utilizes boundary markers by explicitly referencing item-content-trust.md and mandating that all external content be treated as evidence rather than instructions. Its capability inventory includes modifying labels and posting comments via CLI tools. Finally, sanitization and control are enforced through a mandatory human-in-the-loop (HITL) review process where the agent must brief the operator before posing decisions, ensuring the operator remains the final authority.

attend-queue