The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's architecture involves processing untrusted project specifications which are subsequently used to influence the execution of automated tasks.
Ingestion points: references/decomposer-prompt.md ingests {spec_text}; references/curd-prompt.md ingests {prior_handoff} and {spec_summary}.
Boundary markers: The templates interpolate these variables without strong structural delimiters or specific "ignore embedded instructions" warnings for the sub-agent.
Capability inventory: The skill can execute shell commands (via quality_gates), write to the filesystem, and spawn sub-agents.
Sanitization: No explicit sanitization of the specification text or the resulting manifest commands is documented.
[DYNAMIC_EXECUTION]: The skill generates a manifest containing shell commands which are subsequently executed during the workflow.
references/decomposer-prompt.md instructs a sub-agent to populate the quality_gates field in a YAML manifest based on a project specification.
references/wiring-prompt.md specifies a workflow where these quality_gates commands are executed at runtime.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell commands and the management of sub-agent life cycles.
references/manifest-schema.json defines a schema for executing commands in the quality_gates array.
references/spawn-primitive-reference.md details how to render and execute sub-agent phases across various host platforms such as Claude Code, Codex, and OMP.

ultracook