The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted external data during its orchestration phases.
Ingestion points: Phase 0 (Intake) reads external specification and design documents; Phase 1 (Research) utilizes gh search and the Exa search engine to retrieve code and documentation from external repositories and the web (SKILL.md).
Boundary markers: The skill does not explicitly define delimiter-based boundary markers or instructions to ignore embedded commands within the ingested text, though it defines high-level phase transitions.
Capability inventory: The pipeline possesses significant capabilities, including automated implementation via the tdd-guide agent, file system modifications through vertical slicing, and repository commitments (SKILL.md).
Sanitization: There is no mention of sanitization or filtering of the content retrieved from external sources before it is processed by downstream agents.
Mitigation: The risk is significantly reduced by the inclusion of two mandatory human approval gates: Gate 1 (approving the research plan) and Gate 2 (approving the diff and commit messages).
[DATA_EXFILTRATION]: The skill performs network operations to external services as part of its standard research workflow.
Evidence: Phase 1 (Research & Reuse) specifies the use of gh search repos, gh search code, and the Exa search engine to fetch information from external sources (SKILL.md).
Note: These operations are documented as part of the intended development research workflow and target well-known services.

orch-pipeline