project-flow
Skill1 starsChanged 3 months agoMITView on GitHub
Stack-agnostic project orchestration — from discovery through feature-by-feature implementation, with persistent Markdown context and human approval gates. Use when the user wants to start, plan, or resume a project ('start this project', 'plan this project', 'let's build this app'), asks what feature to build next or to implement/prepare/continue a specific feature, or says 'resume from the handoff'. Do not use for isolated conceptual questions, trivial unrelated edits, or pure explanation requests.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- yo1am1/agentic-setuptheir own version, first seen Nov 1, 2026
- loocor/mcpmatetheir own version, first seen Aug 17, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- loocor/mcpmate★ 35same name
- yo1am1/agentic-setup★ 0same name
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 repository4 other files it gives its agents.
- loops-androidSkill
- loops-flutterSkill
- loops-nuxtSkill
- loops-react-nativeSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
21 rules. About 0 are common, 11 are distinctive, and 10 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈11
Distinctive · estimate
This author's own choices. Worth a look.
10
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
Always read existing project documentation before asking questions or making decisions.
- Distinctive: line 25
Do not invoke for: isolated conceptual questions, trivial unrelated edits, or pure explanation requests (e.g.
- Distinctive: line 29
Before anything else, check what exists:
- Very specific: line 36
If docs/plan.md is absent or clearly insufficient (no scope, no phases), this is a new project → go to 1.
- Very specific: line 38
If docs/plan.md, docs/handoff.md, and docs/features/ exist, this is an ongoing project → go to 4.
- Very specific: line 44
Use /grill-me for the initial interview only when substantial product or technical ambiguity remains after reading the repo.
- Very specific: line 46
Ask high-value questions only.
- Very specific: line 47
Use current official documentation (via /research or MCP docs tools) when framework/library versions matter.
- Distinctive: line 48
Stop interviewing as soon as there is enough information for a reliable plan — do not chase completeness for its own sake.
- Distinctive: line 52
Create or update, in this order:
- Distinctive: line 62
After writing plan, handoff, and feature files, STOP.
- Distinctive: line 88
Keep it concise — a roadmap, not a duplicate of the feature files.
- Very specific: line 132
Read, in order: project AGENTS.md/CLAUDE.md, docs/plan.md, docs/handoff.md, the target docs/features/XX-feature.md, then the relevant …
- Very specific: line 134
Do not re-run a full project interview and do not invoke /grill-me merely because a feature is starting.
- Very specific: line 140
Before touching code, add or update an ## Implementation Plan section inside the feature file:
- Distinctive: line 151
Only list meaningful implementation risks or decisions — not routine steps.
- Distinctive: line 168
After successful implementation and validation, update the feature file:
- Very specific: line 171
Add ## Implementation Record — a concise summary of what was built.
- Very specific: line 172
Add ## Validation with checkboxes for what actually ran successfully (lint, typecheck, tests, build, code review, manual verification).
- Distinctive: line 179
STOP — do not start implementing the next feature without approval.
- Distinctive: line 190
Read targeted project files instead of re-running interviews.
≈ “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
project-flow — Project orchestration with persistent context and approval gates
The repository is the source of truth. Persist product decisions, assumptions, architecture decisions, implementation order, feature specs, feature status, handoff context, and validation state in docs/ — never rely on chat history for anything that can live in a file.
Always read existing project documentation before asking questions or making decisions. Never repeat a question the repository already answers.
Activation
Auto-invoke for: starting/planning a project, asking what to build next, asking to implement/prepare/continue a named feature, or resuming from the handoff.
Do not invoke for: isolated conceptual questions, trivial unrelated edits, or pure explanation requests (e.g. "explain how auth currently works").
0. Orient
Before anything else, check what exists:
AGENTS.md/CLAUDE.mdat the project root.docs/plan.mddocs/handoff.mddocs/features/(list files, note numeric order and each## Status)
If docs/plan.md is absent or clearly insufficient (no scope, no phases), this is a new project → go to 1. Discovery.
If docs/plan.md, docs/handoff.md, and docs/features/ exist, this is an ongoing project → go to 4. Starting a feature.
1. Discovery (new or insufficiently planned project)
Reuse anything the repository already states; only ask about genuine gaps.
- Use
/grill-mefor the initial interview only when substantial product or technical ambiguity remains after reading the repo. Do not invoke it mechanically. - Reach for other skills as needed:
/scope,/domain-modeling,/architect,/codebase-design,/research,/prototype. - Ask high-value questions only. Respect explicitly chosen technologies and constraints. Flag contradictions instead of silently picking a side. When the user has no answer, record an assumption in
docs/plan.mdrather than blocking. - Use current official documentation (via
/researchor MCP docs tools) when framework/library versions matter. - Stop interviewing as soon as there is enough information for a reliable plan — do not chase completeness for its own sake.
Write the project docs
Create or update, in this order:
docs/plan.md— see Plan template below.docs/handoff.md— see Handoff template below.docs/features/01-<feature>.md,02-<feature>.md, ... — see Feature file template below. Order by logical dependency and implementation sequence, never arbitrarily.
Initial feature files describe what must be built, not a full implementation plan — that comes later, per feature, in 5. Feature implementation plan.
Stop for global approval
After writing plan, handoff, and feature files, STOP. Report the plan summary and the feature list, and explicitly request approval. Do not write implementation code before the user approves.
Plan template (docs/plan.md)
# Project Plan
## Product Goal
## Users
## Scope
### In Scope
### Out of Scope
## Confirmed Constraints
## Architecture Summary
## Core Domain
## Implementation Phases
- [ ] 01-foundation
- [ ] 02-authentication
- [ ] 03-dashboard
## Feature Dependencies
## Assumptions
## Open Risks
## Definition of Done
## Current StatusKeep it concise — a roadmap, not a duplicate of the feature files.
Feature file template (docs/features/XX-name.md)
# XX - Feature Name
## Status
Planned
## Objective
## User Value
## Scope
## Out of Scope
## Dependencies
## Functional Requirements
## Acceptance Criteria
## Technical Constraints
## Validation Requirements
## Open QuestionsStatus progresses: Planned → Planning → Waiting for approval → Approved → In progress → Implemented → Validated → Completed.
Handoff template (docs/handoff.md)
# Project Handoff
## Current State
## Completed Features
## Current / Next Feature
## Confirmed Decisions
## Important Architecture Context
## Known Issues
## Open Risks
## Files / Areas Worth Reading
## Next Recommended ActionOperational memory for a fresh session or another agent — not a transcript. Update it after meaningful milestones and completed features (use /handoff if useful).
4. Starting a feature (ongoing project)
Read, in order: project AGENTS.md/CLAUDE.md, docs/plan.md, docs/handoff.md, the target docs/features/XX-feature.md, then the relevant existing code.
Do not re-run a full project interview and do not invoke /grill-me merely because a feature is starting.
Only ask the user (or invoke /grill-me) when a material unresolved decision remains after reading the context above — e.g. a data-model change, a permission-model ambiguity, a significant architecture choice, an undefined business rule, or conflicting requirements. Resolve minor implementation details yourself using existing architecture, conventions, and reasonable defaults.
5. Feature implementation plan
Before touching code, add or update an ## Implementation Plan section inside the feature file:
## Implementation Plan
### Steps
### Expected Files / Areas
### Validation Strategy
### Risks / DecisionsOnly list meaningful implementation risks or decisions — not routine steps.
Set ## Status to Waiting for approval and STOP. Do not modify implementation code yet.
Valid approvals: "approved", "go", "ok implement", or an equivalent explicit confirmation. If the user changes the plan, update the feature file and request approval again.
6. After approval — implementation
Delegate to the appropriate skill(s) for the stack and task:
- A stack-specific implementation loop when one exists (e.g.
/loops-nuxtfor Nuxt/Vue,/loops-flutterfor Flutter/Dart,/loops-react-nativefor React Native / Expo,/loops-androidfor native Android) — preferred whenever applicable, since it already owns implementation + validation. - Otherwise compose from
/develop,/implement,/tdd,/test,/debug,/diagnosing-bugs,/check,/code-review,/researchas needed.
project-flow orchestrates the project lifecycle and approval gates; it does not duplicate a stack loop's low-level validation. For the contract a stack loop integration must honor (marker convention, approval gates, safety limit, cleanup, reporting), see docs/loop-integration-contract.md.
7. Feature completion
After successful implementation and validation, update the feature file:
- Advance
## Status(see progression above). - Add
## Implementation Record— a concise summary of what was built. - Add
## Validationwith checkboxes for what actually ran successfully (lint, typecheck, tests, build, code review, manual verification). Never check an item that did not really pass.
Then:
- Mark the feature complete in
docs/plan.md. - Update
docs/handoff.md. - Identify the next logical feature and state it.
- STOP — do not start implementing the next feature without approval.
Approval gates
Two mandatory stops, never bypassed because a next step "seems obvious":
- Global approval — after discovery, once plan/handoff/feature files exist, before any implementation.
- Feature approval — after writing a feature's
## Implementation Plan, before touching its code.
Context efficiency
Read targeted project files instead of re-running interviews. Invoke specialist skills only when they add value, not by default on every feature:
/grill-me— unresolved high-impact ambiguity only./research— external/current technical facts./architect— a significant, undecided architecture choice./prototype— a concrete uncertain technical/design question./domain-modeling— the domain model needs clarification or evolution./handoff— refresh operational context.
Stack-agnostic boundary
project-flow works identically for Nuxt, Flutter, React Native, Android, iOS, backend, or any other stack. It never assumes a specific framework. Stack-specific implementation/validation loops (loops-nuxt, loops-flutter, loops-react-native, loops-android, loops-ios) are separate skills it delegates to in step 6.
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.

