The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the ao command-line utility for goal measurement (measure, validate, drift, history, export, meta, scenarios, render) and status checks (ao status).
[COMMAND_EXECUTION]: The skill runs local shell scripts scripts/validate-output.sh and scripts/validate.sh to perform integrity checks on generated reports and its own configuration.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data sources (source code, artifacts, GOALS.md) which could contain malicious instructions meant to influence the reality check report.
Ingestion points: Reads claims from source documents, repository artifacts, command outcomes, and goal files (GOALS.md or legacy YAML).
Boundary markers: The instructions provide clear constraints to report only facts and gaps, prohibiting the agent from issuing a 'verdict' or 'PASS' and from selecting new work based on findings.
Capability inventory: File system writes to .agents/scratch/ and .agents/ao/goals/baselines/, execution of the ao CLI tool, and execution of local shell scripts.
Sanitization: No explicit sanitization or filtering of evidence content is mentioned before it is included in the report findings.
[DATA_EXPOSURE]: Accesses internal platform metadata and artifact stores located in .agents/ao/intents/ and .agents/ao/verdicts/ to report recency and recency of evidence.

reality-check