The Agent Skills Directory

[SAFE]: The skill implements significant safety mechanisms for automated operation, including a mandatory review sub-agent cycle, test baseline comparisons to prevent regressions, and an accumulation guardrail that pauses the loop if too many minor issues (concerns) build up.
[COMMAND_EXECUTION]: The orchestrator performs version control operations including git add and git commit to manage the implementation lifecycle. It also implicitly runs the project's test suite to establish baselines and verify fixes.
[DATA_EXPOSURE]: The skill reads project-specific configuration and state files (docs/index.json, issues.json, progress-log.md) to maintain state across iterations. This is limited to the local project environment.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it processes untrusted task descriptions from issues.json to drive the behavior of implementation sub-agents.
Ingestion points: docs/<feature-name>/issues.json and docs/index.json (SKILL.md).
Boundary markers: Absent; the skill does not explicitly use delimiters to wrap content from the task list when spawning sub-agents.
Capability inventory: Spawning implementation/review sub-agents, executing git commands, and running the project test suite.
Sanitization: Absent; the orchestrator passes task data directly to sub-agents.
Assessment: The risk is mitigated by the 'automated' mode being restricted to specific implementation sub-agents and the requirement for a separate review sub-agent and test pass before any changes are committed.

simple-run