Skip to content

mattvideoproductions/MVP-Prompts / skills/foreman

foreman

Verified

Be the foreman on a big job. One lead model plans the work, hands pieces to helper agents in lanes that never share files, checks everything before accepting it and reports the cost; or runs a long solo build with an owner in the loop. Covers the North Star plan of record, briefs that carry the lead's reasoning, measured gates plus the owner's judgement, strike-based promotion between model tiers, a cost ledger and a mistake register every pass is checked against. Use it when the user types /foreman, says "be the foreman", "be the lead" or "orchestrate this", asks to delegate to subagents or run lanes or passes, or mentions a North Star, three strikes, tiers or promotion between models. Also load it when a task looks too big for one context or needs several agents or days, even if not asked by name; the skill then has you offer it and wait for a yes. Also use it to resume such a project, close a pass, report a session's cost or record a mistake. Not for a change one agent can finish in one sitting.

references/skeleton.md

View on GitHub
# The skeleton

The skill reduced to the techniques it is built from. Each one is a way of prompting a model, or of arranging the work around it, with the reason it works.

## Give the reader what you know

- **Write for a reader with no history.** A worker has only the brief. Reread it as the only message ever received, and add whatever lives only in the conversation.
- **Pass the reasoning down, not just the request.** State the design and why, the numbers, the failures to expect and what is already ruled out. A list of deliverables leaves the hard decisions to the weaker model.
- **Put the reason beside every decision.** A choice without its reason gets "improved" by the next helpful agent.
- **Quote, do not paraphrase.** The owner's exact words, dated. Point at files and plan sections instead of retyping them.

## Ask in a way that gets the truth back

- **Ask a question; do not hand over a conclusion.** A model given a hypothesis tends to return it confirmed.
- **For an unknown, give suspects and the test that would rule each out.** The worker measures first and tries to disprove one.
- **Give disagreement, guesses and gaps their own lines in the report.** DISAGREE, GUESSED, NOT CHECKED, STILL WRONG. What has a slot gets written.
- **Check with someone who was not told the answer.** A fresh agent gets the work and the test, never the conclusion.

## Define done before the work starts

- **Done is an exact command, a number and a bar.** The bar is fixed first and only the owner moves it.
- **The one being measured does not hold the measure.** Workers cannot edit gates. The lead reruns every one.
- **Test what ships.** Same settings, real input path, real hardware, the copy actually deployed.
- **Numbers, then a look, then a person.** A gate says it is not broken. A look catches what no number measures. The owner says whether it is good.

## Bound the work

- **Scope by ownership.** These files and resources are yours; these you do not touch.
- **Say when to stop and when to give up, in advance.** Stop conditions end a pass. A kill rule ends the task.
- **Fix the shape of the answer.** A report template, short where it can be and long where it matters.
- **Cap the context.** Budget by size. Past it, a fresh agent starts from a short handoff that carries the findings.
- **Cap the whole effort.** A set number of passes, then the best available result with its known gaps.

## Match the model to the task

- **Strongest model on the hardest and least testable work.** Cheapest model that finishes on everything else.
- **Move up on evidence.** Strikes and plateaus are counted, and the allowance shrinks as the price gap to the next tier shrinks.
- **Carry the history up and hand the rest back down.** The stronger model gets what was tried and why it failed. Once the hard part is solved, the remainder returns to a cheaper tier.
- **Blame the brief before the model.** If a stronger model with the same brief would also have failed, fix the brief.

## Keep memory outside the conversation

- **One plan of record, current state first.** Any agent can start from the top of one file.
- **History only grows.** An append-only log, snapshots each pass, and a comparison that flags anything rewritten.
- **Every note from the owner gets an id, a route and a status.**

## Make the system correct itself

- **Meter the cost of every pass.** By tier, by lane, and over time, with the lead's own share shown.
- **Turn each mistake into a record with a check.** What happened, the cause, the rule, and a question that can only be answered by looking.
- **Compare every pass against that record.** The system is tested against its own history, not only its intentions.
- **Promote a lesson that keeps failing.** Prose, then a line in a template, then a script, then the owner. A rule that is written down and still broken needs a stronger form, not a louder sentence.

## Keep the human where only a human works

- **The owner sets the bar and judges the result.** Put something they can run or judge in their hands every pass.
- **Say what nobody has checked.** A green table can hide a product no one has used.
- **Claim only what you just read.** Every fact in a report comes from a file or an output at the time of writing.
- **Guess when it is cheap to undo and flag it. Otherwise ask.**
- **Spend their time last.** Batch what they must test, and count how often they are asked.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall