The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided descriptions and context from the local /foundation/brain.md file to drive its orchestration logic, which creates a theoretical surface for indirect instructions. However, this is mitigated by rigorous confirmation logic and restricted write capabilities.
Ingestion points: Reads initiative details (goals, DRI, metrics) from the user and context from the /foundation/brain.md file.
Boundary markers: The skill uses structured markdown templates for its charter and master document, but does not explicitly instruct the model to ignore embedded data instructions.
Capability inventory: The skill can invoke other agent skills and update specific sections of the local brain context file.
Sanitization: Security is managed through human review; no skill execution or file update occurs without the user explicitly confirming the proposed action (e.g., 'Confirm to start? [Y/N]').
[SAFE]: The skill operates entirely on local context files and does not perform network operations to unknown or untrusted domains. Its functionality is transparent and strictly follows user-approved checkpoints.

workflow-orchestrator