- Home
- Skills
- Agents & Orchestration
- make no mistake
Works with the AI Agents you already use
make no mistake
A rigorous verification loop for high-stakes code, data, and system changes where correctness is non-negotiable.
Free
make no mistake
Example session with this skill installed
I need to update the production database schema to add a unique constraint on user emails. Can I safely run this migration? Make no mistake.
- Read your context and instructions
- Compiled the make no mistake
I have analyzed the migration plan for the users table schema change. Because this touches production data and involves an irreversible schema modification, I have classified this as T1 (Pre-gate audit).
Verification Report
Tier: T1 — local migration audit; no T2 gate crossed (pre-execution)
Outcome: Migration script migrations/004_add_email_unique.sql verified for local safety
Changed: none (verification only)
Verified: main @ d2b8e1a (clean), repo root; psql dry-run against local clone exit 0 (oracle: schema diff matches spec); SELECT count(email) FROM users GROUP BY email HAVING count(*) > 1 returned 0 rows in local staging.
Inferred: Migration will fail in production if concurrent writes insert duplicate emails during the ALTER TABLE execution; reverses if production DB contains existing duplicates not present in staging.
Unverified: Current production duplicate count (requires read-only probe on live environment).
Residual risk: Migration downtime during index build on large table; stopped after verifying logic and local data consistency.
Verifier: none (solo; executed checks verified, unexecuted behavioral claims inferred)
Next steps
- Run the duplicate check query provided in the
Verifiedsection against the live production environment. - Confirm the maintenance window duration for the index build on the
userstable. - Approve the T2 gate to execute the migration.
Connects securely to your tools. The creator never sees your data.
About this skill
make-no-mistake
"Make no mistake" is not a constraint. It is a wish.
make-no-mistake is an agent skill that turns the old vibe-coding incantation into an actual verification loop. It does not promise perfection. It makes the work checkable: define correctness, hunt for specific failure modes, demand observations instead of self-confidence, fix only what is proven broken, and report what remains unknown.
The idea
Most "be careful" prompts add no information. They do not say what a mistake is, which mistakes matter, or how anyone would know. This skill supplies the missing machinery:
- Contract — the goal, observable done criteria, non-goals, assumptions, and stakes (with reversibility checked across technical, data, information, and operational dimensions).
- Mistake map — a pre-mortem across wrong goal, wrong path, wrong facts, wrong change, wrong evidence, and wrong state.
- Predict, then act — write the expected observation before acting; a surprise is a signal.
- Verify — establish provenance and oracle independence, pick the domain-appropriate verification mechanism (software, finance, legal, research, documents, operations), rank evidence, test the test, inspect the diff, check the boundary.
- Correct — confirmed defects only, one fix at a time, with a regression check.
- Fresh eyes — a separate fresh-context verifier (isolated subagent, clean CLI model process, or human) with least-privilege access; for T1 only, an honest solo cold-read with executed checks kept
verifiedand unexecuted behavioral claims markedinferred. - T2 gate — a pre-execution packet covering blast radius and four-dimensional reversibility, pre-state, rollback, abort criteria, post-action checks, approval, and harness-level hardening.
- Report — tier triggers, verified, inferred, unverified, residual risk, verifier, stop reason, and (for T2) ledger, post-action state, and rollback.
It scales ceremony to stakes: T0 uses five questions at the top of SKILL.md, T1 runs the compact core loop, and T2 loads references/t2.md for a persisted risk ledger, a mandatory fresh-context verifier, and — whenever an action crosses the gate boundary — an explicit gate packet and post-action verification.
Why it is not just "try harder"
A model asked to check its own work tends to agree with itself. This skill makes verification external and falsifiable: state the claim, name the observation, make the observation, and distinguish what was verified from what was inferred or assumed. It also has a stopping rule, because endless checking is another failure mode.
Install
The skill package is SKILL.md plus four reference files loaded on demand, with no runtime dependencies. Preserve the make-no-mistake/references/ directory layout when installing or uploading so relative links resolve:
make-no-mistake/
├── SKILL.md
└── references/
├── evidence.md
├── t2.md
├── verifier.md
└── worked-example.md
mkdir -p <skills-dir>/make-no-mistake
cp -r SKILL.md references <skills-dir>/make-no-mistake/
Hardening the gate (operator setup)
A gate written in skill text is soft policy: a miscalibrated shell, interpreter (python, node, curl), or MCP tool call can still cross the gate boundary unless the runtime blocks it. Keep live production credentials unmounted during pre-gate work and enforce the boundary in your harness:
- Codex:
--sandbox(read-onlyorworkspace-write),--ask-for-approval(approval_policy), and.rulesexecpolicy rules. - Claude Code:
permissions(allow/ask/denyin.claude/settings.json) andPreToolUsehooks coveringBashand MCP write/high-impact-read tools. - Cursor: command and MCP allowlists (
permissions.allow/permissions.denyin.cursor/cli.jsonor IDE settings),sandbox(enabledwithnetworkAllowlist), and project or enterprise.cursor/hooks.json(beforeShellExecution,beforeMCPExecution,preToolUsewithfailClosed: true). Note that user-level~/.cursor/hooks.jsondoes not reach Cloud Agents, and CLI hooks can fail to fire even withfailClosed: true.
Smoke-test the gate before relying on it: trigger a harmless forbidden sentinel command in a throwaway sandbox and confirm it is actually denied before trusting hooks or allowlists around production. (Checked 2026-11-17 against codex-cli 0.167, Claude Code 2.1, and Cursor CLI 2026.11; flags and hook behavior change — verify against current vendor documentation and <cli> --help.)
Use
See the description frontmatter in SKILL.md for the authoritative trigger rules, or invoke /make-no-mistake explicitly when your harness supports slash-skill invocation.
Examples:
- "Add rate limiting to the login endpoint. Make no mistake."
- "Here is my migration plan. Double-check the order before I run it."
- "The test passed but the behavior is still wrong. How do you know the fix works?"
Honest limits
- It cannot guarantee zero mistakes. It replaces "trust me" with "here is the evidence and the residual risk."
- It costs time and tokens. Use it where being wrong is expensive, not as ceremony on every edit.
- A separate verifier helps, but the same model can share the same blind spots. Agreement is not verification.
- It does not replace tests, type checks, code review, or human judgment.
- Prompt rules alone are soft policy: T2 verifier requirements, waivers, and execution gates are defined in
references/t2.md; back them with the smoke-tested harness controls above.
The name
"Make no mistake" became a joke about magical prompting: a huge, underspecified request plus an imaginary --no-bugs flag. thesysdev/make-no-mistakes is the satirical M-Stack version. This project is the sincere inverse: not a magic flag, a loop.
The name is a joke. The loop is not.
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 human gate approval for T2 actions. Effectiveness depends on access to independent oracles or test suites. Cannot self-waive T2 rigor.
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 today
- Free with an Agensi account
- Software
Git