The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data, specifically specification and design documents, to drive the 'Intake' and 'MVP inner loop' phases. This creates a potential surface for indirect prompt injection if the processed documents contain malicious instructions intended to influence the agent's planning or implementation steps.
Ingestion points: Phase 0 (Intake) reads specification/design documents to extract scope and feature lists; Phase 1 (Research) utilizes external search tools (gh search, Exa).
Boundary markers: The skill explicitly implements a 'gated, not autonomous' architecture. Gate 1 requires human approval of the task list before implementation, and Gate 2 requires human approval of the diff before committing.
Capability inventory: The pipeline has the capability to write implementation code (Phase 4), execute tests via a TDD workflow, and perform conventional commits (Phase 6).
Sanitization: No specific technical sanitization or escaping of input data is mentioned, relying instead on human oversight at defined gates.
[COMMAND_EXECUTION]: The orchestrator delegates tasks to various platform-level commands and specialized agents (e.g., /feature-dev, /plan, /code-review, /gan-build). These operations are consistent with the skill's primary purpose as a developer orchestration engine and are managed through human-in-the-loop gates.

orch-pipeline