The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill orchestrates work based on instructions found in documentation files.
Ingestion points: Reads job descriptions and state from docs/01-START/roadmap.md and docs/graphs/*.md.
Boundary markers: Employs structured sections like "Alignment Contract" and "Checklist".
Capability inventory: Can execute shell commands and spawn subagents via the Agent tool.
Sanitization: Explicitly requires human approval for the initial graph and mandates 'human gates' for all sensitive or high-risk actions.
[DYNAMIC_EXECUTION]: Constructs prompts for subagents using data retrieved from local graph files. This is a standard agentic pattern for parallel task execution and is constrained by the mandatory "alignment contract" passed to each subagent.
[COMMAND_EXECUTION]: The skill uses the date command for recording activity in the change log, which is a benign administrative function.

graph-running