Project Dev Workflow
Development discipline for AI coding agents: confirm the request, layer and verify, adversarial review, pitfall logging
Free
Works with the AI Agents you already use
Project Dev Workflow
Example session with this skill installed
Add a rate limiter to the public API and make sure existing clients don't break.
- Read your context and instructions
- Compiled the project dev workflow
Restates the request as a baseline (goal, deliverables, scope, constraints, assumptions, decisions needed) and waits for confirmation. Then splits it into verifiable layers — config surface, middleware, per-route limits, client-facing headers, load test — runs each layer's check before starting the next, has an independent verifier confirm the final layer against the criteria, and closes with a report listing evidence, unverified items, and any new PITFALLS.md entries.
Connects securely to your tools. The creator never sees your data.
About this skill
A working discipline for AI coding agents on non-trivial development work. It fixes four recurring failure modes: coding on a misread request, declaring work done without running it, defending a design from a single viewpoint, and repeating a pitfall that was already solved.
How it works:
- Tiering — small, unambiguous, reversible tasks take a light path with no ceremony; everything else runs the full workflow. A light task that grows is upgraded mid-task.
- Requirement baseline — restate the request as goal, deliverables, scope in/out, constraints, acceptance criteria and assumptions; act only after confirmation. Timeout is not consent.
- Layered plan — split into layers, each with criteria checkable by a command or observation; unfalsifiable criteria like "good quality" are rejected.
- Verification — verify each layer before the next. An independent verifier gets the baseline, criteria, changed files and diff, never the implementer's reasoning; verdicts are pass / fail / cannot verify with evidence. Two failed fixes on one criterion means stop and find the root cause.
- Evidence rule — nothing is called done or passing unless it was run. "Implemented, not verified" is the allowed phrasing.
- Adversarial review — critic subagent, then multi-perspective reviewers, then explicit self-refutation and pre-mortem when subagents are unavailable. Every critique is accepted with a change or rejected with a reason.
- Pitfall logging — PITFALLS.md committed for anything any machine would hit; a machine-level log for anything specific to one machine. Entries indexed past ~150 lines, obsolete entries archived.
- Proactive execution — improvements inside the confirmed scope are made and reported; anything outside it, behaviour-changing or irreversible is proposed with value and cost.
- Autonomous mode — on "just do it", decisions are made from the recommended option and recorded as assumptions. Data deletion, deploys, pushes, migrations and paid API calls are still held for the user.
Instruction-only: no scripts, hooks, or network calls. Works with any agent that loads SKILL.md.
Instruction-only: the rules take effect only if the host model follows them; nothing is enforced by code. Without subagent support the skill falls back to self-verification and self-refutation, which are weaker than an independent verifier. SKILL.md is about 19 KB and loads in full on each trigger, which matters under a tight context budget. No benchmarks or evaluation results are published, so claims about its effect would be unsupported.
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 today
- Free with an Agensi account
Requires
Permissions
File Scopes
No network access. No scripts or hooks; every rule is instruction text. The skill reads and writes PITFALLS.md in the project it is used on, and may propose a one-line addition to AGENTS.md / CLAUDE.md.