- Home
- Skills
- Code Quality & Review
- SpecGate - Spec governance for AI coding agents
More screenshots
Works with the AI Agents you already use
SpecGate - Spec governance for AI coding agents
Give coding agents approved context, check delivery evidence against each acceptance criterion, and keep human acceptance in control. Free, local-first skills for spec-driven development.
Free
SpecGate - Spec governance for AI coding agents
Example session with this skill installed
Use SpecGate in an isolated Local demo to verify the criterion “Unhealthy requests return 503.” Use the reviewed Node JUnit command and exact selected testcase. Show what happens when the runner exits 0 but that case is missing, then add it and rerun with the same Context Pack. Stop before human acceptance.
- Read your context and instructions
- Compiled the specgate - spec
Delivery review: a green runner still had an evidence gap
This walkthrough is reproducible with the shipped selected-test guide. The CLI and Node runner were executed in a fresh Local fixture; no existing project store was used.
First observation
The command exited 0, but its JUnit report contained only “healthy returns 200.” The agreed “unhealthy returns 503” testcase was absent.
| Acceptance criterion | Runner exit | SpecGate outcome | Reason |
|---|---|---|---|
| Unhealthy requests return 503 | 0 | Unmet; submission exit 1 | Selected testcase is missing |
State: implementation. Next actor: implementing_agent. The observed failure reason was retained in delivery evidence rather than treated as successful coverage.
Repair and rerun
I added the exact unhealthy testcase and updated the completion evidence. The same work item, Context Pack and reviewed verification pin were retained. A fresh submission produced new completion and review records.
| Observation | Result |
|---|---|
| Selected testcase | test / unhealthy returns 503 / passed |
| Check provenance | selected_test_observed |
| Submission exit | 0 |
| Context Pack | Unchanged |
| Watched test file | Changed against the pin; warning retained |
| Final state | awaiting_acceptance |
| Next actor | human_reviewer |
Human handoff
Review the changed test and retained watched-input warning, then decide against the exact review and acceptance basis. No acceptance command was executed. A selected passed case does not establish runner honesty, semantic completeness or continuous immutability.
Complete walkthrough: https://github.com/thanhtung2693/specgate/blob/main/docs/using-specgate/guides/observe-selected-tests.md
Connects securely to your tools. The creator never sees your data.
About this skill
SpecGate is a free, open-source skill pack for solo developers and teams building software with coding agents. It connects approved intent to reviewable delivery evidence.
What it helps you do
- Hand an agent one approved Context Pack with clear scope, criteria and non-goals.
- Review claims, check outcomes, citations and evidence gaps per acceptance criterion.
- Optionally pin reviewed commands, exact JUnit selectors and watched inputs in Local mode.
- Resume with checkpoints and inspect changes since the previous handoff.
- Keep human approval and acceptance explicit, with evidence and review history retained.
Four focused skills
The pack contains the SpecGate router, project setup, work preparation and work delivery skills, plus the preservation reference. These are the canonical instructions shipped with SpecGate.
Fits your existing workflow
Keep authoring in OpenSpec, Spec Kit, Superpowers, Markdown or your own format. Your authoring tool keeps its files and lifecycle. REA and other coding agents can continue planning and implementing; SpecGate supplies the approved handoff, verification evidence and acceptance record through its CLI.
Getting started
Install the SpecGate CLI separately using the linked documentation, initialize Local mode, and use the supported integration for Codex, Claude Code or Cursor. Keep the four skill folders as siblings. Use one installation owner per IDE; avoid duplicate copies alongside an existing native or CLI-managed plugin.
Local mode needs no Docker, server, connected account or model API key. Full mode is optional and should run on a trusted machine or private network.
Trust and limits
A green command can omit a required testcase. Selected-test observation can expose that gap, but does not prove runner honesty or test quality. Commands run in your host shell without a sandbox. Readiness and passing evidence do not replace human approval or acceptance.
Free to use and share under Apache-2.0. The ZIP contains instructions, not the CLI executable.
Requires a separately installed SpecGate CLI. Verification commands run unsandboxed on the host and must be reviewed. Passing checks or selected JUnit cases do not prove test quality or runner honesty. Human approval/acceptance remain authoritative. Full mode belongs on a trusted machine or private network. Use one installation owner per IDE.
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 any payment.
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 forever
No account required to browse
Trust & safety
Security scan
Security scan passed 22 days ago
- Free with an Agensi account
Requires
Permissions
File Scopes
Reads selected project specs and source. Writes CLI-owned .specgate evidence and approved implementation files. Setup may update selected IDE integration files with user approval. Network is used for installation and optional Full server calls. Local use needs no connected service account. Environment variables select CLI configuration/workspace; credentials must not be printed.