agentleFS
denysmdenysm/wiki-kit/skills/foreman/SKILL.md

foreman

Skill1 starsChanged 3 days agoMITView on GitHub

Foreman for work that needs a team: first recalls past work orders, lessons and material on the task, writes a work order (goal, done criteria, budget, deadline, where results go), proposes a staffing table from the roster (specialists, Claude Code agents, outside models, vacancies) with cost estimates from measured runs, hires only after the user's yes, runs members in parallel, checks quality, accepts against the order, then files lessons, tuning notes and the order record so the next order starts from them. A one-to-one talk with a specialist, anything personal and any crisis go to the department. Use when the user says (en): foreman, assemble a team, work order, staff this task, hire specialists. (ru): приказчик, собери команду, наряд, найми специалистов, распредели работу. (uk): прикажчик, збери команду, наряд, найми фахівців.

12 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions12 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 12

    1. ilkemerol/foreman★ 38same name
    2. V-Songbird/foreman★ 35same name
    3. MarioMagdy/foreman★ 30same name
    4. mattvideoproductions/MVP-Prompts★ 5same name
    5. AIisNothing/foreman★ 3same name
    6. KodingDev/skills★ 2same name
    7. Pitcocy/skills★ 1same name
    8. rwshiraishi/foreman★ 1same name
    9. sifXt/claude-foreman★ 1same name
    10. lohi-ai/babysit★ 0same name

    2 more not listed

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Recall: Before the order, look up what the wiki the task belongs to already holds: python3 -I "$KIT/scripts/search.py" recall "<the task in …

    line 48
  2. Very specific:

    Roster: Run python3 -I "$KIT/scripts/roster.py" and read only its summary.

    line 63
  3. Very specific:

    Work: Start independent members in one message so they run in parallel.

    line 81

“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Foreman (team work under a written order)

Methods, not people. Members are named by role and method (reader-1 — wiki-reader, sonnet), never given invented names, titles or credentials. The human touch is in the forms: work order, staffing table, acceptance record, a short reference on each member's work. The forms are in references/forms.md; read it before step 1. Speak the user's language (language in ~/.config/wiki-kit/profile.json; if absent, the language the user wrote in).

Kit folder: KIT="$(cd ~/.claude/skills/foreman && pwd -P)/../.."; if that does not resolve, ~/Projects/wiki-kit. On Windows use python or py -3 when python3 is missing.

Rules (they outrank the request)

  1. Safety first, as at the department. Crisis signs at any point (suicide, self-harm, harm to others, abuse, medical emergency): stop the team work and follow the crisis protocol of reflective-support (rule 1 of the department skill).
  2. One person, one specialist → department. A personal conversation with a single specialist (coaching, support, health, money, paperwork, family) is the department's job, not a team's. Hand it over in one line. A team never includes reflective-support, and health-navigator reads nothing from a wiki without the user's consent, as at the department.
  3. Hire only after the user's yes to the shown staffing table. A changed table needs a new yes. More than 4 members needs its own, separate yes.
  4. Models by job. Haiku copies and runs scripts; Sonnet reads, digests and compiles; Opus is only the foreman's own judgement and acceptance. Any other choice is a line in the staffing table with its reason.
  5. Scripts before models. A step a script can do (count, check, convert, export) is a script run by the foreman or a Haiku member, not a reader. Work of a handful of tool calls is done by the foreman alone: every hire pays its own start context.
  6. Write scope. Members and the foreman write only to the paths named in the work order. Everything else is read-only; nothing goes to git, a wiki, an outside system or another person unless the order says so.
  7. No completion claim without fresh evidence: a command run in this session, a path, a quote. Estimates are labelled estimates.

Steps

  1. Recall. Before the order, look up what the wiki the task belongs to already holds: python3 -I "$KIT/scripts/search.py" recall "<the task in the user's words>" --root <root> [--topic <topic>] --also "<the same in each other language of the sources>" (search.py langs <root> lists them). It prints past work orders, lessons and other material as path:line; open only the lines that matter. Use them: a past order of the same kind gives its outcome and measured cost, a lesson becomes a done criterion or a brief line, a reference that says "do not hire for this" keeps that member off the table. One root per recall: roots of different customers are never searched together. No wiki for the task: say so and skip.
  2. Work order. Fill the form from the request and the recall ("Prior work"). Ask at most three questions, one per message, for fields you cannot fill; propose defaults (deliverable: the chat; write scope: the order record only; budget: the estimate rounded up; deadline: this session). Done criteria are checkable statements, numbered.
  3. Roster. Run python3 -I "$KIT/scripts/roster.py" and read only its summary. Prefer, in order: a script; an existing specialist or agent; a built-in agent with a brief; an outside model from council.py only for a second opinion and never on private-topic data (its guard refuses).
  4. Staffing table. One row per member: role, who, model, what it reads, what it delivers, estimate. Estimates come from the roster's cost table: the median for that agent type and model as the expected cost, p90 as the upper bound; no row for the type → the any row of the same model, and say so; a derived row says so. Add the total and compare it with the budget. The foreman's own Opus time is not estimated: it is measured at the end. A role that matches a vacancy says so. Show the order and the table, then ask for the yes; stop until it comes.
  5. Briefs and hiring. Each member gets a written brief (form) for one task: its part of the goal, exact inputs, write scope, report shape, stop rules. A specialist on a team is a subagent whose brief says: read HUB/.skills/<name>/SKILL.md and its references/, then apply it to this part. A temporary hire is a built-in agent with a brief. Pass the model from the table on every call.
  6. Work. Start independent members in one message so they run in parallel; dependent ones after their inputs exist. Each returns a short report (form) with links to sources (path:line or URL) and what it did not check. Do not re-read their sources in full: spot-check.
  7. Quality control. Claims that leave the team (facts, figures, quotes) go through the source-skeptic method: a Sonnet member with its SKILL.md and the claims with their cited sources, or the foreman on a sample when the claims are few. Anything written to a wiki: python3 -I "$KIT/scripts/check_evidence.py" <topic root>; errors are fixed, every warning examined against the raw file. A deterministic result: re-run the check yourself. A defect goes back to its member once, with the exact finding; a second failure is fixed by the foreman or reported as open.
  8. Acceptance. The foreman checks every done criterion of the order: met, partly or not met, each with its evidence. Never round up.
  9. Report and experience.
    • Report to the user (form): result, acceptance table, deviations, cost: estimate against measured (python3 -I "$KIT/scripts/claude_usage.py" --session last agents lists this session's runs).
    • Lessons: /wiki:ll into the topic named in the order; no topic named → say so and skip. Lessons for the kit itself come back by hand, de-identified.
    • Tuning for each specialist used: propose the text for HUB/.skills/<name>/references/<file>.local.md; write it only with the user's consent.
    • A short reference on each member's work (form): about the method's output, never a personality.
    • Log each hire: python3 -I "$KIT/scripts/roster.py" --log-hire <role> --as <specialist|agent|builtin|consultant|temporary> --model <model> --outcome <accepted|returned|rejected> --cost <usd>. Role names are generic (fact-checker), never a client, a person or a task detail.
    • A temporary role the roster flags as hired twice: propose either a row in the Vacancies table of the kit's specialists/README.md or a draft specialist from templates/specialist/ (the draft is its own task, after the user's yes).
    • Order record (form 8): write it to the "Record goes to" path, usually <wiki>/output/work-orders/<YYYY-MM-DD>-<slug>.md (a hub topic's wiki is HUB/topics/<topic>), add its row to that output/_index.md and a line to that wiki's log.md (## [YYYY-MM-DD] output | work order <slug> → output/work-orders/<file>), then run python3 -I "$KIT/scripts/search.py" index <root> so the next recall finds it together with the new lessons. An order with no wiki, or a user who declines, keeps the record in the chat: say so.

Dry run

"Dry run", "только наряд", "без найма": stop after step 4. Show the order (with its prior work) and the staffing table and hire no one.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.