The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project documentation and code (e.g., discovery.md, proposal.md, specs/**/*.md, design.md, tasks.md, align.md, and test output) as grounding artifacts for its decision-making process. This represents an ingestion surface where malicious instructions embedded in these files could attempt to influence the agent. However, the risk is minimized by the skill's specific reasoning constraints and its exclusion of security-critical items.
Ingestion points: Project artifacts and code files specified in the grounding requirements.
Boundary markers: None explicitly defined within the skill instructions to delimit external content.
Capability inventory: The agent is instructed to read files and run tests (file system and shell access).
Sanitization: No explicit sanitization or validation of the ingested content is mentioned.
[SAFE]: The skill incorporates several security and quality best practices for autonomous agents:
Isolation: It requires a fresh subagent for each resolution to prevent reasoning bias or context contamination from the main session.
Auditability: All autonomous decisions, including the claims, opposing arguments, and residual risk assessments, are logged to dark.md for human review.
Safety Override: The skill explicitly forbids autonomous resolution of security requirements or mitigations, ensuring that security-critical decisions always involve a human user.

hrt-dark-mode-decision-gate