- Home
- Skills
- Agents & Orchestration
- Lean Execution: Verified, No-Fluff Agent Output
More screenshots
Works with the AI Agents you already use
Lean Execution: Verified, No-Fluff Agent Output
Generic "be concise" prompts make it worse: agents start skipping tests and hiding uncertainty.
$4
Lean Execution: Verified, No-Fluff Agent Output
Example session with this skill installed
CI is red on main. Fix it.
- Read your context and instructions
- Compiled the lean execution
Summary
Illustrative session. The agent jumps to the first failing CI step, reproduces it locally, fixes the root cause (no test disabled) and reports in four lines.
What the agent did (no narration shown to the user)
| Step | Action |
|---|---|
| Inspect | Read only the first failing step of the CI log and its stack trace |
| Reproduce | Ran npm run test:e2e -- checkout locally: same failure |
| Diagnose | checkout.spec.ts used a hard-coded date that has now expired |
| Fix | Replaced the hard-coded date with a fixed test clock |
| Verify | Re-ran the failing step locally: pass |
Report delivered
DONE: Fixed checkout.e2e — expired hard-coded date replaced by a fixed clock (tests/e2e/checkout.spec.ts:14).
VERIFIED: npm run test:e2e -- checkout (local) — pass. CI re-run not triggered from here.
BLOCKED: NONE
NEXT: Push to re-run CI.
What it did NOT do
- Paste the 900-line CI log
- Announce a plan or narrate each command
- Disable the failing test to turn CI green
- Ask "should I continue?"
Next steps
- Push the branch to re-run the pipeline.
- Run the included
evals/tasks/04-failing-ci.mdchecklist on your own repo to compare runs with and without the skill.
File names, commands and results in this demo are illustrative.
Connects securely to your tools. The creator never sees your data.
About this skill
Coding agents spend a large share of their output on things you did not ask for: restating your prompt, announcing obvious plans, narrating every tool call, reading whole files for one line, recapping finished work and asking "should I continue?". Generic "be concise" prompts make it worse: agents start skipping tests and hiding uncertainty.
Lean Execution removes the waste and keeps the substance.
What it does
- Acts instead of narrating: no plan announcements for obvious work, no tool commentary.
- Retrieves narrowly: search before read, line ranges instead of whole files, reuse of prior output.
- Decides without micro-confirmations: proceeds on reversible decisions; asks one precise question only for irreversible, destructive or unknowable ones.
- Verifies by risk: a 10-row risk matrix from typo fixes to production deploys, each with a minimum check.
- Reports in four lines: DONE / VERIFIED / BLOCKED / NEXT, where VERIFIED names the actual check run.
- Holds a quality boundary: never skips required tests, never invents results, never hides partial work.
What's inside (27 files)
SKILL.md: core rules, activation phrases, risk matrix, ask-or-proceed rule, report formatreferences/: anti-patterns with bad vs. good examples, quality boundary, verification playbook, retrieval playbook, modes guidetemplates/: ready-to-paste rules for Claude Code (CLAUDE.md), Cursor (.mdc and .cursorrules), Codex CLI (AGENTS.md), GitHub Copilot, plus a one-line session starterevals/: 10 realistic test tasks (bug fix, refactor, failing CI, risky migration, auth change...) with pass/fail checklists, and a benchmark protocol to measure tokens and turns with vs. without the skill on your own workexamples/: 3 illustrative before/after transcripts
Modes
lean (default) · deep exhaustive analysis · status result/blocker/next only · continue resume without recap
Install in 30 seconds
Claude Code: drop the folder in ~/.claude/skills/. Cursor: copy templates/lean-execution.mdc to .cursor/rules/. Codex CLI: append the AGENTS.md snippet. Copilot: append the copilot-instructions snippet.
Measured results (A/B test, Claude Code + Sonnet, 18 sessions)
| Without skill | With skill | |
|---|---|---|
| Tasks solved correctly | 9/9 | 9/9 |
| DONE / VERIFIED / BLOCKED / NEXT report | 0/9 | 7/9 |
| Regression test added (failing-test task) | 0/3 | 2/3 |
| Output tokens (median) | baseline | +19% to +50% |
Honest note: on short tasks with a concise modern model, the skill did not reduce tokens; it adds verification work. Its measured value is verification discipline and predictable reports.
Minimum sufficient context. Minimum necessary narration. Maximum useful execution. No quality sacrifice.
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.
Does not reduce tokens on short tasks with concise modern models: in our A/B test (Claude Code + Sonnet) it used 19-50% more output tokens due to added verification. Long multi-step sessions and agents other than Claude Code are not yet benchmarked. It is a behavior skill: results depend on the agent following its instructions (structured report adopted in 7/9 runs).
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
- Software
- vitest
Node.js
Git