The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains an 'OWNER OPERATING DIRECTIVE' using highly assertive language ('ABSOLUTE', 'NON-NEGOTIABLE', 'THIS PREAMBLE WINS') designed to override other instructions and force the agent into a specific Engineering Manager persona. It also mandates the use of 'ultrathink' authority for all subagents.
[DYNAMIC_EXECUTION]: The skill's primary workflow involves spawning subagents to generate code (Staff Engineer) and then executing that code via test runners (npx vitest) and build tools. This creates a loop where AI-generated content is executed on the local system.
[EXTERNAL_DOWNLOADS]: The skill includes an 'Auto-recovery protocol' that performs git clone operations on multiple external GitHub repositories at runtime to establish a 'Platform Evidence Base'. These fetches from GitHub target various repositories including those from OpenAI, Google Gemini, and Microsoft.
[COMMAND_EXECUTION]: The skill heavily utilizes system commands via the gh CLI, npm, and git to manage issues, PRs, and releases. These commands are executed based on inputs retrieved from external GitHub issues and PRs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub (issue bodies, PR diffs, comments) and uses it to drive agent orchestration and code generation. The lack of explicit output sanitization before data is passed to subagents creates a potential attack surface.
Ingestion points: Data enters through gh issue view, gh pr view, and gh pr diff in triage-issue.md and review-pr.md.
Boundary markers: None identified in the prompt interpolation logic.
Capability inventory: Significant system access including ctx_execute, ctx_batch_execute, npm, git, and gh tools.
Sanitization: No automated sanitization is described; the skill relies on manual verification by subagents (e.g., 'Claim Verification' and 'Git Archaeology').

context-mode-ops