- Home
- Skills
- Agents & Orchestration
- Gungnir Evidence Gate
Works with the AI Agents you already use
Gungnir Evidence Gate
Lock the goal. Verify real artifacts and command exit codes. Repair failed work before an AI coding agent says "done.
Free
Gungnir Evidence Gate
Example session with this skill installed
Verify that the new logging utility in src/utils/logger.ts is implemented and the existing unit tests pass.
- Read your context and instructions
- Compiled the gungnir evidence gate
I have executed the Gungnir Evidence Gate to verify the requested logging utility implementation. The task has reached a
PASS verdict based on the deterministic evidence collected from the workspace.
Delivery Verdict
Objective: Implement logging utility and verify via unit tests.
Status: PASS
| ID | Criterion | Type | Result | Evidence |
|---|---|---|---|---|
| AC-1 | src/utils/logger.ts exists | Artifact | PASS | File located at /project/src/utils/logger.ts |
| AC-2 | npm test -- src/utils/logger.test.ts | Command | PASS | Exit code: 0 |
| AC-3 | Logger exports info, warn, error | Content | PASS | Grep confirmed required symbols in source |
Executed Checks
- Checked for artifact existence at specified path.
- Executed test runner for the specific module.
- Inspected file content for required interface exports.
Next steps
- Review the generated
src/utils/logger.tsfor any style preference adjustments. - The DispatchContract JSON has been removed from the temporary directory.
- You can now proceed to integrate the logger into the main application entry point.
Connects securely to your tools. The creator never sees your data.
About this skill
Coding agents often hallucinate success, claiming a task is done when tests are still failing or files are missing. This skill forces agents to verify work against real workspace evidence before declaring completion. It replaces "I think I'm done" with a deterministic gate based on exit codes and artifact checks.
What it does
- Lock acceptance criteria by converting user requirements into a formal, observable DispatchContract JSON.
- Execute deterministic verification using the Gungnir CLI to validate command results and file states.
- Categorize delivery verdicts into explicit Pass, Fail, Partial, or Unverifiable states based on ground truth.
- Automate repair loops by identifying specific failed criteria and forcing targeted fixes until the gate clears.
- Generate evidence reports that document exactly what was checked, which commands ran, and what artifacts exist.
How it works
- Goal Locking: Define the objective and map requirements to observable command exit codes or file paths in a temporary contract.
- Work Execution: The agent performs the requested coding task using its standard toolset.
- Verification: Run
npx dsh-gungnir-verifyagainst the workspace and contract to get a model-free verdict. - Failure Reconciliation: If the gate returns a Fail or Partial status, analyze the mismatch and apply targeted repairs.
- Final Reporting: Present the verdict, executed checks, and any unverifiable obligations to the user.
Frameworks & tools
This skill requires Node.js to run the dsh-gungnir-verify package via npx. It is designed for coding agents with shell and file system access like Claude Code, Cursor, and Codex.
Why this beats prompting it yourself
Standard prompts rely on the LLM to "double-check" its own work, which frequently leads to confirmation bias. This skill offloads verification to a deterministic CLI, ensuring that an agent cannot claim success unless the environment actually satisfies the contract.
Use cases
- Gate PR ready status on successful test suite execution and linting exit codes.
- Verify the existence and content of specific generated artifacts after a refactor.
- Prevent false positives in autonomous coding loops by requiring external evidence.
- Document compliance with acceptance criteria for complex multi-step migrations.
Known limitations
Requires a Node.js environment to execute the verifier. It cannot verify requirements labeled as sandbox-external, such as external API states or live URL deployments.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires Node.js for CLI verification. Cannot verify 'sandbox-external' requirements. Does not replace the agent loop; requires tool access to execute commands and read files.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 15 days ago
- Free with an Agensi account
- Software
- dsh-gungnir-verify
Node.js