idiot-boss
Skill2 starsChanged 38 days agoMITView on GitHub
Run the idiot-boss orchestration approach when a cheaper coordinator handles logistics and light QA while stronger models own planning, design, and technical judgment.
In this repository22 other files it gives its agents.
- atddSkill
- automated-checksSkill
- bddSkill
- behavior-specificationSkill
- choose-orchestrationSkill
- codebase-designSkill
- code-reviewSkill
- compile-instructionSkill
- computer-use-wizardSkill
- gherkinSkill
- goal-loopSkill
- goal-superviseSkill
- high-value-testingSkill
- hvtddSkill
- retroSkill
- setup-engineering-skillsSkill
- tddSkill
- test-implementationSkill
- testingSkill
- ubuntu-desktop-computer-useSkill
- ubuntu-machine-configurationSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 32
Before retrying a disappointing result, inspect the prompt the worker actually received.
- Distinctive: line 34
If the same failure returns after a repair, route the evidence to a strong model for diagnosis or replanning.
- Distinctive: line 36
Before reporting on a worker, confirm its state from the host's lifecycle evidence.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Idiot-boss orchestration
The coordinator owns glue and light QA. Strong models own the decisions and do the substantive work directly. In design, the desired taste lives in the designer's weights: preserve its room to interpret the brief.
Establish the plan
Use the existing plan supplied by the user or invoking workflow, including its model assignments, input boundaries, and completion criteria. For ambitious work without a plan, give a strong model the original request and source artifacts. Ask it to produce the steps, dependencies, and observable completion criteria the coordinator needs to follow the plan.
Give the coordinator explicit operational instructions: which step to dispatch next, its inputs, how completion is checked, and when to return to the planner. A complex plan or hard-to-test product warrants a more capable coordinator; use the selected orchestration configuration.
Dispatch the work
For each ready step, send its strong worker:
- The user's original brief and relevant source artifacts, including target screenshots or specs.
- The step's objective, dependencies, and constraints from the user or plan.
- The expected artifact and evidence of completion.
Frame the problem and desired outcome. Leave design and technical solutions to the worker. Preserve user requirements verbatim where paraphrasing could change them; give the worker direct access to the sources instead of a chain of summaries.
The coordinator follows the plan, dispatches ready work, tracks dependencies and artifacts, and runs the prescribed checks. Questions requiring design, architecture, technical tradeoffs, or changes to the plan go to the responsible strong model with the relevant evidence. The coordinator relays its decision and continues.
Check and repair
Use observable evidence for light QA: prescribed commands and their results, missing artifacts, broken interactions, or visible overlap. Send failures and screenshots back as observations against the brief. The strong worker chooses the repair. Subjective design quality and subtle correctness questions go to a strong evaluator.
Before retrying a disappointing result, inspect the prompt the worker actually received. Remove solutions invented by the coordinator and restore missing source material. Keep decisions and updated artifacts available directly to subsequent workers so the goal survives handoffs.
If the same failure returns after a repair, route the evidence to a strong model for diagnosis or replanning. Report a concrete unresolved blocker when progress requires unavailable input or capability.
Before reporting on a worker, confirm its state from the host's lifecycle evidence. When a worker stops without a recap, such as after an error or usage limit, report what happened.
Finish when the plan's completion criteria are met. Report the artifacts and verification evidence, including any evaluation that remains unresolved.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

