The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local script, scripts/graph-audit, to validate the structural integrity of files modified during the vision definition process. This script acts as a linter for the workspace and is part of the skill's mandatory safety filing gate.
[INDIRECT_PROMPT_INJECTION]: The skill reads from wiki/business/vision-stub.md to populate candidate strategy fields. This represents an indirect injection surface, which is managed through internal safety protocols and human oversight.
Ingestion points: The skill reads input from the local stub file wiki/business/vision-stub.md in Step 1.
Boundary markers: The skill requires a manual 'checkpoint' (Step 3) where the user must explicitly confirm the text before it is written, and it uses a 'durable claim admission gate' to verify factual statements.
Capability inventory: The skill writes to several documentation files (wiki/business/vision.md, index.md, hub/business.md, log.md, baton.md) and has the ability to execute the local scripts/graph-audit utility.
Sanitization: Data is sanitized through the 'shared durable claim admission gate' which requires all durable claims to be verified against calculations or quoted receipts, and the scripts/graph-audit tool ensures files follow established structural conventions.

vision