AgentSkeptic
VerifiedVerify agent tool effects against actual stored state instead of traces.
What is AgentSkeptic?
AgentSkeptic is an open-source verification agent that validates whether an AI workflow actually modified the expected records in databases and other stores. It reads tool events and a registry of expected outcomes, then compares them to live state rather than relying on execution traces alone.
The system executes a check command that returns machine-readable verdicts on stderr and a structured Outcome Certificate on stdout. Results indicate trusted, not_trusted, or unknown status based on whether observed state matches registry expectations for SQL, HTTP, object storage, and similar targets.
It targets teams shipping agent-driven workflows who need reliable evidence before irreversible actions such as billing or deployment. Users maintain a tools registry in version control and emit events that the checker evaluates against configured stores.
Capabilities
What you can build with AgentSkeptic
Pre-release workflow validation
Run checks before shipping or billing to confirm that claimed database changes actually occurred.
CI pipeline integration
Add state verification steps to continuous integration so failed state matches block merges.
Audit trail for agent actions
Generate Outcome Certificates that document whether downstream records matched expectations after each run.
Install AgentSkeptic
npx agentskeptic check --workflow-id wf_example --project ./path/to/your-app --db ./path/to/readable.sqlitenpx agentskeptic check --workflow-id wf_example \
--project ./path/to/your-app \
--db ./path/to/readable.sqlite- 1Install via npx so the agentskeptic binary is available without global setup.
- 2Create or update agentskeptic/tools.json in your project to map tool IDs to verification rules.
- 3Emit workflow events to events.ndjson or pass them explicitly during the check.
- 4Execute the check command with workflow ID, project path, and database connection.
- 5Review the Outcome Certificate on stdout and the truth_check_verdict lines on stderr.
AgentSkeptic: pros & cons
Pros
- +Provides deterministic certificates based on actual stored state rather than logs.
- +Runs entirely locally with no API key or license server required for core checks.
- +Supports multiple store types through a registry-driven contract model.
- +Integrates into existing CI and development workflows with simple CLI flags.
Cons
- –Requires maintaining a tools registry and event emission in the target application.
- –Advanced features such as drift detection and baselines require a paid enforce mode.
- –Initial setup involves configuring readable database access and event paths.
Frequently asked questions
No, the default check command runs locally without any license or key.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar agents
Other automation options worth comparing.
