The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts (scripts/doctor.py and scripts/validate_plan.py) to verify environment configuration and lint plan manifests. The doctor.py script performs read-only inspection of sensitive local configuration files (e.g., ~/.codex/config.toml) and interacts with a local loopback router to verify model availability.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external repository data and structured plan manifests, creating a surface for indirect prompt injection.
Ingestion points: The skill ingests untrusted data from project repository files during discovery, user-supplied build requests, and structured plan.json manifests.
Boundary markers: The workflow includes a dedicated review phase (references/review.md) where the orchestrator must perform a multi-lens review of worker results against the original specification and baseline.
Capability inventory: Capabilities include local Python script execution, file system modifications in targeted directories (docs/agent-work/), and native subagent delegation to worker models.
Sanitization: scripts/validate_plan.py provides structural linting, detects template placeholders, and enforces path validation to prevent traversal outside the repository root (rejecting .. and .git paths).

astra-flash-orchestrator