The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes a local Python zipapp (skills/mold/scripts/mold.pyz) to perform artifact resolution, spec validation, and system finalization. It also starts a local HTTP server for the review canvas feature using python3 skills/mold/scripts/mold.pyz review serve.
[DYNAMIC_EXECUTION]: Through the 'Prototype Cycle' defined in references/prototype-cycle.md, the skill is instructed to spawn isolated sub-agents to 'build a throwaway' code prototype. This is intended to settle design unknowns by executing generated logic in a hermetic environment.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests data from repository wikis via Hallouminate and external research via Briesearch (Tavily).
Ingestion points: Data enters the context during the 'Grounding' phase (references/grounding.md) and 'Validate Cycles' (references/validate-cycle.md).
Boundary markers: The skill employs a 'Two-key handshake' checklist (references/handshake.md) and a 'gate-graph' state machine (references/gate-graph.md) to ensure coherence.
Capability inventory: The skill can write durable specification artifacts, ADRs, and issues to the filesystem (references/curdle.md) and can initiate the /cook command for implementation.
Sanitization: Sanitization is handled via an 'Agent-introduced scope' audit table and a 'Goal coverage' check that requires the agent to trace all new nouns and goals back to the original user input.
[EXTERNAL_DOWNLOADS]: The skill uses the researcher agent role and the /briesearch tool to fetch external research data from Tavily to validate design hypotheses.

mold