The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive shell operations using git, gh (GitHub CLI), jq, and a specialized knowledge-gate CLI. These tools are used for repository management (branching, committing, pushing), workflow automation, and metadata extraction. While the commands appear bounded to the skill's operational logic, shell execution remains a primary capability.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection due to its core function of ingesting and processing untrusted data from merged Pull Requests.
Ingestion points: Content, titles, and evidence from merged Pull Requests discovered via the GitHub MCP (SKILL.md, Step 1 and Step 3).
Boundary markers: None are explicitly defined at the orchestrator level to delimit untrusted PR data from instructions.
Capability inventory: The orchestrator can spawn sub-agents, write to the filesystem, commit and push to the repository, modify GitHub labels, and trigger GitHub Actions workflows (SKILL.md, Step 3, 4, and 7).
Sanitization: The skill does not describe explicit sanitization or validation of the ingested PR content before passing it to sub-agents or including it in generated reports.
[COMMAND_EXECUTION]: The skill utilizes gh workflow run to perform a 'Graceful Handoff', triggering a new instance of the GitHub Actions workflow to circumvent token expiration. This represents an automated control of the platform's CI/CD environment.

batch-refine