The Agent Skills Directory

[SAFE]: The skill implements a highly structured derivation chain for project planning, moving through Goal Mirroring, Context Research, Decisions, Requirements, and Task generation. This methodical approach reduces the risk of automated errors or unexpected behaviors.
[SAFE]: It utilizes mandatory human-in-the-loop (HITL) approval gates using the AskUserQuestion tool at Layer 2 (Decisions), Layer 3 (Requirements), and Layer 4 (Tasks). This ensures the user retains control over the generated content.
[SAFE]: Local codebase analysis is performed using standard tools (Grep, Glob, Read) and internal sub-agents. No external network requests to non-whitelisted or untrusted domains were found.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection during the L1 Context Research phase, as it reads codebase files that could contain malicious instructions designed to influence the agent's planning logic.
Ingestion points: Codebase files are scanned via Glob, Grep, and Read in the references/L0-L1-context.md file.
Boundary markers: None explicitly defined for research data, though results are formatted into a structured ## Research section.
Capability inventory: The skill can write local files (Write), create directories (Bash), and invoke sub-agents (Agent).
Sanitization: No explicit sanitization or filtering of the research findings is mentioned before being processed by the orchestrator or sub-agents.
Mitigation: The impact is significantly mitigated by the three mandatory human approval gates (L2, L3, L4) where the user reviews the derived decisions, requirements, and tasks.

specify