The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local bash and node scripts to retrieve configuration and validate state transitions.
Evidence: Invocations of bash "${CLAUDE_PLUGIN_ROOT}/scripts/get-config" to read settings.
Evidence: Execution of node scripts/guard/guard.mjs to perform state-change guard checks.
Evidence: Use of a local CLI tool ssf for retrieving and setting workflow state.
[PROMPT_INJECTION]: The skill processes untrusted content from project files to drive its internal logic, creating a surface for indirect prompt injection.
Ingestion points: The skill reads content from proposal.md, specs/*.md, design.md, tasks.md, and execution-contract.md across multiple files.
Boundary markers: No explicit delimiters or safety instructions are provided to the agent to distinguish between file data and control instructions when parsing these documents.
Capability inventory: The skill has the capability to execute local scripts and modify the environment's workflow state based on the findings from these files.
Sanitization: There is no evidence of sanitization or strict schema validation performed on the markdown content before it is used to determine the next skill in the workflow.

workflow-orchestrator