agentleFS
Luis85Luis85/agentic-workflow/.claude/skills/project-run/SKILL.md

project-run

Skill0 starsChanged 6 months agoMITView on GitHub

Triggers: "let's start a client project", "set up a service engagement", "I need project management for this work", "start a project for [client]", "manage this as a client engagement", "I have a client brief" AskUserQuestion works only in the main thread. Subagents cannot ask the user anything, so all clarification happens in this conductor turn before dispatching or between project phases. project-run is the conversational entry point for the Project Manager Track — the opt-in layer for service-provider teams delivering…

The file

Show its rules and text

We split this skill into the rules it gives an agent

22 rules. About 0 are common, 5 are distinctive, and 17 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 17

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask the user to describe the engagement in one paragraph: who is the client, what are they getting, when, and at what cost (if known).

    line 40
  2. Distinctive:

    Ask for a project slug (engagement-level name, not solution name).

    line 42
  3. Distinctive:

    Ask the user to share the client brief, statement of work, or scope description.

    line 47
  4. Very specific:

    Run /project:initiate — the project-manager agent will draft the four management documents and produce a go/no-go decision brief.

    line 48
  5. Very specific:

    Ask: "Are there existing features (specs/) to link, or do we need to start new ones?"

    line 54
  6. Very specific:

    If yes: record the specs/<slug>/ paths in project-description.md's deliverables section.

    line 55
  7. Very specific:

    If no: recommend /spec:start <feature-slug> for each deliverable, then /project:initiate after to link them.

    line 56
  8. Distinctive:

    Ask: "Do you have a known client communication cadence (weekly, bi-weekly)?

    line 57
  9. Very specific:

    Run /project:close — the agent will draft the closure document and ask for sponsor sign-off.

    line 62
  10. Distinctive:

    If the user mentions a new client request, out-of-scope work, or an unexpected scope change:

    line 67
  11. Very specific:

    Before each phase, read projects/<slug>/project-state.md:

    line 74
  12. Very specific:

    If phase: scaffolded → Phase 1 (initiation pending)

    line 75
  13. Very specific:

    If phase: executing → Phase 2 (check if weekly log is current)

    line 76
  14. Very specific:

    If phase: closing → Phase 3 (closure in progress)

    line 77
  15. Very specific:

    If phase: closed → remind user about /project:post

    line 78
  16. Very specific:

    If no project-state.md exists: run Phase 0 + Phase 1.

    line 80
  17. Very specific:

    Never do project-manager work in your own turn.

    line 86
  18. Very specific:

    Never call AskUserQuestion from inside a subagent prompt.

    line 87
  19. Very specific:

    Never ask more than one AskUserQuestion per gate.

    line 88
  20. Very specific:

    Use the project-specific state model in projects/<slug>/project-state.md: phase, initiation_status, closure_status, and last_updated.

    line 89
  21. Very specific:

    Never write directly to projects/<slug>/ during normal phase execution.

    line 90
  22. Very specific:

    If project-manager returns "blocked — needs human input", surface its question to the user via a single AskUserQuestion, capture the …

    line 94

≈ “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.

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

project-run — Project Manager Track conductor

Triggers: "let's start a client project", "set up a service engagement", "I need project management for this work", "start a project for [client]", "manage this as a client engagement", "I have a client brief"

AskUserQuestion works only in the main thread. Subagents cannot ask the user anything, so all clarification happens in this conductor turn before dispatching or between project phases.


What this skill does

project-run is the conversational entry point for the Project Manager Track — the opt-in layer for service-provider teams delivering software to clients. It guides the user through the full project lifecycle: initiation → execution (weekly rhythm) → closure → post-project evaluation.

It sequences specialist commands (/project:*) and gates with AskUserQuestion at every human-decision point (go/no-go, change approval, sign-off).

This skill does not drive Specorator feature work (/spec:*) or Discovery Track sprints (/discovery:*) — it coordinates them from the project management layer.


When to run

Run when:

  • A client engagement is starting and governance is needed.
  • The user says they have a statement of work, contract, or client brief.
  • Multiple features need to be coordinated under a single delivery commitment.
  • Scope, schedule, budget, or stakeholder reporting are concerns.

Do not run when:

  • The user is building an internal product with no client.
  • There is no contract boundary.
  • The user just wants to run a feature — recommend /spec:start instead.
  • The user has a blank page and needs ideation first — recommend /discovery:start instead.

Procedure

Phase 0 — Orientation

  1. Intake gate. Before asking the user to describe the engagement, list inputs/ non-recursively and surface every item via a single AskUserQuestion: "I see N items in inputs/. Which are relevant for this engagement?" If inputs/ is empty, print one line ("inputs/ is empty — no source material to consult") and proceed. Never auto-extract zips or archives — extraction is a separate, explicitly approved step. Cite paths into inputs/ from project-description.md and other canonical artifacts. Full contract: docs/inputs-ingestion.md. Decision: ADR-0017.
  2. Ask the user to describe the engagement in one paragraph: who is the client, what are they getting, when, and at what cost (if known).
  3. Confirm the user wants the full project management layer (not just a Specorator feature). If they're uncertain, explain the difference (see docs/project-track.md §1).
  4. Ask for a project slug (engagement-level name, not solution name). Suggest corrections if needed.

Phase 1 — Bootstrap and Initiation

  1. Run /project:start <slug>.
  2. Ask the user to share the client brief, statement of work, or scope description. Accept any format (prose, email, bullet list, doc path).
  3. Run /project:initiate — the project-manager agent will draft the four management documents and produce a go/no-go decision brief.
  4. Present the go/no-go brief. Ask the user (acting as sponsor): "Do you approve initiating this project?" Gate on approval before advancing.

Phase 2 — Execution rhythm

  1. Recommend: "Run /project:weekly each week. I'll read all linked feature workflow-state.md files and produce a weekly log entry with RAG status."
  2. Ask: "Are there existing features (specs/) to link, or do we need to start new ones?"
    • If yes: record the specs/<slug>/ paths in project-description.md's deliverables section.
    • If no: recommend /spec:start <feature-slug> for each deliverable, then /project:initiate after to link them.
  3. Ask: "Do you have a known client communication cadence (weekly, bi-weekly)? I'll include focused-comms notes in the weekly log."

Phase 3 — Closure (when user signals project end)

  1. Confirm all deliverables have been accepted or explicitly deferred.
  2. Run /project:close — the agent will draft the closure document and ask for sponsor sign-off.
  3. Recommend: "Set a reminder to run /project:post in 3–6 months to evaluate whether the expected benefits were realised."

At any point — change control

If the user mentions a new client request, out-of-scope work, or an unexpected scope change:

  • Prompt: "This sounds like a change request. Want me to run /project:change to log and assess it?"

State reading

Before each phase, read projects/<slug>/project-state.md:

  • If phase: scaffolded → Phase 1 (initiation pending)
  • If phase: executing → Phase 2 (check if weekly log is current)
  • If phase: closing → Phase 3 (closure in progress)
  • If phase: closed → remind user about /project:post

If no project-state.md exists: run Phase 0 + Phase 1.


Constraints

  • Never do project-manager work in your own turn. Drafting the project description, registers, reports, closure, or post-project evaluation belongs to /project:* commands and the project-manager agent.
  • Never call AskUserQuestion from inside a subagent prompt.
  • Never ask more than one AskUserQuestion per gate. Batch choices into a single question.
  • Use the project-specific state model in projects/<slug>/project-state.md: phase, initiation_status, closure_status, and last_updated. Do not use generic status: active | paused | blocked transitions for this track.
  • Never write directly to projects/<slug>/ during normal phase execution. The /project:* commands and project-manager own those files; this conductor reads state, asks the user, and dispatches.

When a project phase escalates

If project-manager returns "blocked — needs human input", surface its question to the user via a single AskUserQuestion, capture the answer, and re-dispatch the same /project:* command with the answer as additional context. Do not answer on the user's behalf.


References

  • Methodology: docs/project-track.md
  • ADR: docs/adr/0008-add-project-manager-track.md
  • Agent: .claude/agents/project-manager.md
  • Commands: .claude/commands/project/ (start, initiate, weekly, change, report, close, post)
  • Templates: templates/project-state-template.md + templates/project-description-template.md + templates/deliverables-map-template.md + templates/followup-register-template.md + templates/health-register-template.md

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.