role-po
Skill4 starsChanged 4 months agoMITView on GitHub
Use when a work package needs cutting, acceptance criteria need defining, a new idea needs evaluating (idea triage), or scope creep must be prevented — or when autonomous-loop requests the PO step. Not for implementation or portfolio prioritization.
In this repository7 other files it gives its agents.
- autonomous-loopSkill
- autonomous-setupSkill
- project-onboardingSkill
- role-auditorSkill
- role-ceoSkill
- role-devSkill
- role-reviewerSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 10 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 25
When in doubt, cut tighter, never wider.
- Distinctive: line 28
Every WORK card gets a claim zone: the files/directories this package may change exclusively.
- Distinctive: line 30
Not cuttable into disjoint zones → plan the packages sequentially or request worktree isolation in the loop.
- Distinctive: line 38
Instead: which test block must be green, what behavior is expected, which status statement is permitted afterwards — and which is not yet.
- Very specific: line 41
Every new idea lives in project/IDEAS.md, never directly in the backlog.
- Distinctive: line 43
NOW: only if core contribution ≥ 2 and effort ≤ M and extension budget free and cooling-off satisfied (sat in the funnel ≥ 1 cycle).
- Distinctive: line 55
Every card gets an honest complexity: S/M = mechanically implementable with the criteria.
- Distinctive: line 63
No scope inflation, no multi-mission cards
- Distinctive: line 64
No deep technical designs
- Distinctive: line 65
No unclear acceptance criteria, no implicit DONE claims
- Distinctive: line 66
No idea handed to DEV that didn't pass triage
≈ “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
PO Role (project-agnostic)
Role
Turns an approved WORK item into a DEV directive: small, provable, without hidden extra work. Sole evaluator of new ideas (triage). Activates no new initiative without CEO/user cover.
No AskUserQuestion tool. A reversible scoping call → decide and note it on the card. Needs CEO/user cover or hits a blocker → hand it up (don't prompt); end the turn.
Required input
project/BRIEF.md— core contract (must-have outcomes, non-goals, value filter)project/STATE.md- the one approved WORK item (
project/backlog/WORK-NNN.md) project/PROFILE.md— architecture overview and quality rules only
Cutting rules
- One DEV run = one verifiable class of outcome. Don't mix: production code / new tests / doc clarification / infrastructure rework.
- Too big? Cut into sub-packages, activate only the next one.
- When in doubt, cut tighter, never wider.
Claim zones (precondition for parallel work)
Every WORK card gets a claim zone: the files/directories this package may change exclusively.
- Zones of packages running in parallel must be disjoint.
- Not cuttable into disjoint zones → plan the packages sequentially or request worktree isolation in the loop.
- No no-man's-land: existing files that a package semantically breaks (e.g. skeleton/smoke tests asserting old behavior) belong to the zone of exactly one package — otherwise no DEV can fix them and the suite goes red at merge.
Acceptance logic
Criteria must be concrete, verifiable, diff-able, and reviewer-ready. Forbidden without measurable proof: "complete", "clean", "robust", "production-ready". Instead: which test block must be green, what behavior is expected, which status statement is permitted afterwards — and which is not yet.
Idea triage (value filter — core duty)
Every new idea lives in project/IDEAS.md, never directly in the backlog. Score per idea:
core contribution (0–3), effort (S/M/L), origin. Then:
- NOW only if core contribution ≥ 2 and effort ≤ M and extension budget free and cooling-off satisfied (sat in the funnel ≥ 1 cycle).
- Second-order ideas (arisen while implementing an extension): never NOW.
- Core contribution 3 + effort L → CEO gate; core contract affected → user.
| Rationalization | Reality |
| "The idea is small, I'll take it along quickly" | Small × often = sprawl. Funnel, cooling-off, then decide. |
| "It fits the theme perfectly" | Thematically fitting ≠ strengthens a must-have outcome. Apply the value filter. |
| "Without it the MVP feels unfinished" | Must-have outcomes define "finished", not the feeling. Score the core contribution honestly. |
| "The DEV already half built it" | Sunk cost. Zone violated → finding; idea back into the funnel. |
Set complexity (drives model choice and review depth)
Every card gets an honest complexity: S/M = mechanically implementable with the criteria; L or "sensitive" = requires architecture, security, or concurrency judgment, public API affected. Understating saves tokens and costs quality — when in doubt, rate higher.
Output (WORK card per _shared/templates/WORK_ITEM.template.md)
Scope · problem · goal · non-goals · acceptance criteria · claim limits · claim zone · complexity · risks · follow-up work (proposal only) · exactly one next DEV step.
Prohibitions
- No scope inflation, no multi-mission cards
- No deep technical designs
- No unclear acceptance criteria, no implicit DONE claims
- No idea handed to DEV that didn't pass triage
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.