The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or unauthorized data access were detected in the provided files. The skill's behavior is consistent with its stated purpose as an orchestrator.
[PROMPT_INJECTION]: The skill processes untrusted natural language input from the user to categorize tasks and generate plans for other agents (Indirect Prompt Injection surface), but includes mitigation steps.
Ingestion points: User descriptions of evolution tasks provided during the Phase 1 Intent Interview as described in SKILL.md.
Boundary markers: No explicit delimiters or boundary instructions are defined for user input interpolation in the skill body.
Capability inventory: The skill is configured with access to the Bash, Read, and Write tools in the SKILL.md frontmatter.
Sanitization: The skill incorporates an 'evals HARD-GATE' to prevent the use of unvalidated output and explicitly instructs the agent to confirm tool availability with the user before dispatching work.

os-architect