The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Markdown content from a wiki directory to generate answers and syntheses, creating a surface for instructions embedded in data to influence agent behavior.
Ingestion points: Markdown files representing concepts, governance policies, and indices within the wiki/ directory.
Boundary markers: Instructions require the agent to categorize all claims into evidence classes (Fact, Interpretation, Contradiction, Inference) and apply transparent trust tiers.
Capability inventory: The agent can write synthesis files to the wiki/synthesis/ directory using a bundled helper script.
Sanitization: All writes require a multi-step human approval process. Files are validated for schema and policy compliance by a local script before being committed, and the writer script performs explicit path confinement checks to prevent traversal.
[COMMAND_EXECUTION]: The synthesis persistence script utilizes subprocess.run to execute a local validator utility.
Evidence: scripts/save_synthesis.py executes scripts/validate_governance.py via the system Python executable to audit staged changes for structural errors and broken links before finalizing the write operation. This execution is limited to the bundled validator script.

okf-wiki-query