- Home
- Skills
- Agents & Orchestration
- Spec-Driven Dev for Coding Agents
Works with the AI Agents you already use
Spec-Driven Dev for Coding Agents
Write the spec set (constitution, EARS requirements, design, small tasks), gate it READY/REVISE/BLOCK before code, drive your agent task by task, then check the result for drift. Spec Kit and Kiro compatible.
$14
Spec-Driven Dev for Coding Agents
Example session with this skill installed
Here's our spec folder for login hardening (requirements.md, design.md, tasks.md) and our constitution. Gate it before we let Claude Code implement it.
- Read your context and instructions
- Compiled the spec-driven dev
Spec gate: login hardening → BLOCK
1 Blocker · 2 Major · 1 Minor. Fix B-01 before any code.
| ID | Severity | Location | Finding | Fix |
|---|---|---|---|---|
| B-01 | Blocker | design.md §4 | Design uses JWT sessions; constitution P4 requires cookie sessions for web clients | Switch to httpOnly cookie sessions, or amend P4 via an ADR first |
| M-01 | Major | requirements.md R1.1 | "respond fast" has no threshold | WHEN a user submits the login form THE SYSTEM SHALL respond within 300 ms at p95 |
| M-02 | Major | design.md §4 | No rollback for the lockout-table migration | Add a down-migration and a task T0 to test it |
| m-01 | Minor | tasks.md T2 | References R9.9, which doesn't exist | Change to R2.1 |
Traceability (trace_check.py specs/login --tests tests)
| ID | In tasks | In tests |
|---|---|---|
| R1.1 | NO | NO |
| R1.2 | yes | yes |
| R2.1 | yes | NO |
| N1 | NO | NO |
Orphan task: T3 "Refactor logging" has no requirement. Drop it or add the requirement it serves.
EARS rewrite
- Before: "The system should handle failed logins securely."
- R1.2 IF 5 failed logins occur for one account within 60 s THEN THE SYSTEM SHALL return 429 for 15 minutes.
- R2.1 WHILE the account is locked THE SYSTEM SHALL show the unlock time.
First task brief (once READY)
T1 Add rate limiting to POST /login
Satisfies: R1.2 Constitution: P2, P4
Don't: change session format, add dependencies
Done when: pytest tests/auth/test_rate_limit.py passes;
6th attempt in 60 s returns 429
Stop and ask if: middleware can't be extended without breaking R1.x
Next: apply the fixes (I can edit the spec files), re-run the gate, then start T1.
Connects securely to your tools. The creator never sees your data.
About this skill
Coding agents are fast and confident, and they fill every gap in your instructions with a guess. Spec-driven development removes the guesses before code exists: decide what you're building, write it down in a form the agent can't misread, break it into tasks small enough to review, then check the result against the spec.
Most spec tools do one piece of this. This kit runs the whole loop in four modes.
1. Write the spec set
- Constitution: 8–15 testable project-wide principles (e.g. "no new runtime dependency without an ADR").
- requirements.md: user stories with acceptance criteria in EARS notation (WHEN … THE SYSTEM SHALL …), numbers instead of adjectives, plus explicit out-of-scope and open-questions lists.
- design.md: interfaces, data model and migrations, error handling, security, observability, alternatives, risks and a rollback plan.
- tasks.md: ordered tasks of half a day or less, each mapped to requirement IDs with a done-check (a test or command that proves it).
2. Gate it before any code
Checks for contradictions (against the constitution, ADRs and existing code), coverage (every requirement has a task, every task a requirement), testability ("fast", "secure" and "robust" get flagged), omissions (error paths, auth, rollback, rate limits, privacy, accessibility), task size and order, and stale references. Findings are rated Blocker/Major/Minor with fixes and a READY / REVISE / BLOCK verdict.
3. Drive the agent
One task brief at a time, covering what to do, what not to touch, the done-check, and when to stop and ask. The rule: spec changes go into the spec first, then the code.
4. Verify for drift
Traceability from requirement to code to test. It catches scope creep, silently dropped requirements, design deviations without an ADR, and weakened tests. The verdict is MATCHES SPEC / DRIFT FOUND / INCOMPLETE. The included trace_check.py cross-checks requirement IDs across requirements, tasks and test files, and flags vague wording.
Fits your setup
Works standalone, and uses your existing GitHub Spec Kit or Kiro layout if you have one, so you never run two spec systems. One line in AGENTS.md / CLAUDE.md makes every agent session read the specs first.
It never invents requirements; unknowns become open questions. For a one-line bug fix it scales down to a 3-line spec.
It plans, gates and verifies, but doesn't replace code review or tests. Quality depends on the inputs: it won't invent requirements, so gaps become open questions. trace_check.py matches requirement IDs (R1.2, N3), so tests and tasks need to reference them. Spec Kit and Kiro file names change between versions, so it follows what's actually in your repo.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Permissions
Read Files reads your repo's specs, docs and tests. Write Files creates and updates spec files (requirements.md, design.md, tasks.md) when you ask it to. Terminal/Shell runs the included trace_check.py (Python standard library only, no network calls).