The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to manage the project's file structure and check for environment-specific configuration overlays.
Evidence: [ -n "$POWERSUIT_OVERLAY" ] && ls -1 "$POWERSUIT_OVERLAY/red-team" 2>/dev/null || true used to check for local instruction overrides.
Evidence: mkdir -p red-team && cp -n "${CLAUDE_PLUGIN_ROOT}/templates/red-team/plan.md" red-team/plan.md used to initialize the red-teaming plan file.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input describing AI feature behavior and policies to generate a test plan. This content is persisted in a local markdown file, which could serve as an injection surface if the data contains malicious instructions intended for downstream red-teaming tools.
Ingestion points: User input defining the AI feature, system prompts, and policies within SKILL.md.
Boundary markers: Absent. The skill writes user-provided content directly into the red-team/plan.md file.
Capability inventory: Directory creation (mkdir), file management (cp), and shell execution for environment checks.
Sanitization: No explicit sanitization or validation of the user-provided feature descriptions or policies is performed before persistence.

pressure-plan