The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs system shells to run git and gh commands for automated repository and tracker management in scripts such as verify-push-state.sh, gh-stage.js, and tracker-issue.js. Command arguments are primarily constructed from validated data and repo-derived IDs. A centralized access gate in defer-mutation.js intercepts and redacts sensitive operations.
[DYNAMIC_EXECUTION]: The qa-execute-snippets.mjs script extracts and runs bash code blocks from markdown documentation files to verify accuracy. Execution is sandboxed using a minimal environment, a command allow-list (SAFE_COMMANDS), and regex-based filters to block dangerous patterns like unauthorized redirections.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface where content from repository documents is ingested by AI agents.
Ingestion points: Agents defined in finalise-dod-ac-prompt.md and finalise-dod-security-prompt.md read content from work item files and pull request diffs.
Boundary markers: Present; the system uses structural Markdown delimiters and specific fenced block info strings.
Capability inventory: Subprocess execution is present in qa-execute-snippets.mjs (spawnSync), and tracker mutations are performed in tracker-issue.js and tracker-comment.js (execFileSync).
Sanitization: Present; implementation utilizes strict command allow-lists for shell execution and a centralized mutation gate with secret redaction for tracker interactions.

finalise