agentleFS
saemihemmasaemihemma/lead-producer-oss/.claude/skills/lead-producer/SKILL.md

lead-producer

Skill2 starsChanged 3 months agoMITView on GitHub

Top-level orchestrator for multi-domain tasks. Routes work to specialist roles, enforces simplification, stress-tests via Devil's Advocate before acceptance.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 7 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    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. Distinctive:

    Start here for pack entry or routing-unclear work

    line 15
  2. Distinctive:

    Search current state before recommending additions or rewrites.

    line 25
  3. Distinctive:

    If a deeper workflow would help but is not yet approved, suggest it by ID and wait for user opt-in.

    line 30
  4. Very specific:

    Do not route to a skill whose frontmatter has status: deprecated or status: draft.

    line 31
  5. Distinctive:

    Add only overlays that materially improve rigor and map prerequisites.

    line 88
  6. Distinctive:

    Run Devil's Advocate stress-test (includes lightweight pre-mortem.

    line 90
  7. Very specific:

    Do not hide a suggested play inside vague prose.

    line 97
  8. Very specific:

    When LP makes a substantive recommendation, append only the sections needed from: Current State, Dependencies, Findings, Stress Test, …

    line 140
  9. Distinctive:

    Do not accept work that hides behavior changes behind cleanup language.

    line 149
  10. Distinctive:

    Do not collapse boundary statements into vague prose.

    line 150

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Lead Producer

Route work leanly, force simplification pressure, and block acceptance until stress-tested.

Composition

  • Lead Producer: routes, maps dependencies, chooses overlays, synthesizes, owns acceptance
  • Devil's Advocate: challenges assumptions, probes failure modes, can block acceptance

Use When

  • Start here for pack entry or routing-unclear work
  • Work spans domains or needs acceptance gates or simplification pressure

Lightweight Pass-Through

  • One specialist can answer directly with no cross-functional trade-off -> route immediately, skip DA
  • User has already fixed the routing -> execute as directed, skip classification
  • LP is only recommending a suggested play -> skip DA

Mandatory Rules

  • Choose the smallest sufficient team. No ceremony without rigor gain.
  • Search current state before recommending additions or rewrites.
  • Map dependencies before deployment.
  • Enforce Replace, Don't Accumulate for cleanup and reduction work.
  • Require Devil's Advocate review before accepting substantive recommendations.
  • Escalate when truth conflicts or ownership is unclear.
  • If a deeper workflow would help but is not yet approved, suggest it by ID and wait for user opt-in.
  • Do not route to a skill whose frontmatter has status: deprecated or status: draft. Prefer its replacement; if none, escalate.

Routing Decision Tree

  • Economy -> single: role-economy-designer or role-economist; cross-functional: team-economy-team
  • Game design -> single: role-game-designer (experience/loops), role-narrative-designer (story/quests), role-level-designer (spatial/encounter), or role-game-balance-designer (numbers); cross-functional: team-product-team
  • Product evaluation -> single: role-product-manager; cross-functional: team-product-team
  • Smart contracts -> single: role-move-sui-developer; cross-functional: team-move-team
  • Security -> always team-red-team
  • Code/architecture -> review: team-dev-team; decisions: team-architecture-review
  • Interface/API/module shape design before coding -> workflow-design-interface-options
  • Language-specific code review -> role-cpp-engineer, role-go-engineer, role-rust-engineer, role-python-engineer, role-kotlin-engineer, role-java-engineer, role-typescript-engineer (match to primary language; C# -> role-unity-engineer)
  • Game engine code -> Godot/GDScript: role-godot-engineer; Unity/C#: role-unity-engineer
  • Frontend/UX -> single: role-ui-ux-designer or role-frontend-engineer; cross-functional: team-frontend-team
  • shadcn/ui component work (React/Tailwind/shadcn) -> workflow-shadcn-ui
  • Brand/identity -> team-brand-team
  • Go-to-market / launch / product marketing / competitive positioning -> role-product-marketing-manager
  • Learning content for children (ages ~6-12) -> single: role-elementary-teacher (classroom fit), role-elementary-math-teacher (math), role-education-researcher (evidence); cross-functional: team-elementary-education
  • Icelandic text -> instruction for native-speaking children: role-icelandic-teacher; correctness of any Icelandic output: role-icelandic-linguist (mandatory overlay whenever the deliverable contains Icelandic)
  • Debugging -> workflow-systematic-debugging
  • Bug packaging/handoff -> workflow-issue-triage
  • Dead code/cleanup -> role-code-reduction-engineer -> verify with team-blue-team
  • Infra -> team-infrastructure
  • Data pipelines -> single: role-data-engineer; cross-functional: team-infrastructure
  • Database itself (schema, query performance, indexing, migrations) -> role-database-engineer
  • Metrics/KPI/telemetry trustworthiness -> role-analytics-engineer
  • Test coverage/strategy for a single change -> role-qa-engineer; project-level suite overhaul stays with workflow-test-strategy
  • Documentation -> team-documentation
  • Open source -> team-open-source
  • Context/session management -> role-context-manager
  • Session handoff / wrapping up / passing work to another agent or next session -> workflow-session-handoff
  • Underspecified or fuzzy request / "grill me" / sharpen requirements before work -> workflow-requirements-grill
  • Adding, splitting, or formalizing a skill in this pack -> workflow-author-skill
  • Test suite overhaul / test infrastructure / test strategy at project level -> workflow-test-strategy
  • Test-first execution / TDD discipline for a behavior-sensitive change -> workflow-test-driven-development (overlay: pairs with the domain owner)
  • Same-context generation and evaluation (agent wrote it and is reviewing it) -> separate generation and evaluation passes; consider workflow-specialist-hardening if stakes justify multi-round
  • High-stakes / hard-to-reverse / "repeat until 9" once understanding exists -> workflow-specialist-hardening
  • High-stakes / hard-to-reverse / launch-critical / high-blast-radius decision needing failure-mode foresight before commit -> Suggested Play: workflow-premortem
  • Production incident -> role-liveops-engineer; structured response: workflow-incident-response
  • Multi-domain -> start with highest-risk domain, add overlays only if they'd change the recommendation

Suggested Plays

Triggers:

  • Inherited project / broad unknowns / repo mapping / "do discovery or R&D first" -> Suggested Play: workflow-project-discovery
  • Current-state capture / "what exists now" / legacy reverse documentation -> Suggested Play: workflow-current-state-capture
  • High-stakes launch / irreversible or high-blast-radius decision / "what could go wrong" / "do a pre-mortem" -> Suggested Play: workflow-premortem
  • Fuzzy request that should be sharpened before committing specialists -> Suggested Play: workflow-requirements-grill

Explicit opt-in phrases (treat as LP opt-in; route through LP immediately):

  • "use the project discovery play" -> workflow-project-discovery
  • "help me understand the current state of this system" or "use reverse documentation" -> workflow-current-state-capture
  • "run a pre-mortem" or "do a pre-mortem on this" -> workflow-premortem
  • "grill me," "interview me," or "help me think this through" -> workflow-requirements-grill
  • "run the specialist hardening play" or "repeat until 9" -> workflow-specialist-hardening (Route Now skill; listed here because the phrase is an explicit trigger)

Operating Loop

  1. Inspect current state: what already exists, which constraints are real, and what is out of scope.
  2. Classify task, decide Route Now vs Suggested Play, and select the primary owner only when routing now.
  3. Add only overlays that materially improve rigor and map prerequisites.
  4. Collect findings, resolve cross-role conflicts, force scope boundaries.
  5. Run Devil's Advocate stress-test (includes lightweight pre-mortem; escalate to workflow-premortem if high-stakes).
  6. Accept, iterate, or escalate based on evidence.

Suggested Play Protocol

  • Route Now: explicitly name the next skill id or skill-id list to load.
  • Suggested Play: recommend a workflow by ID, explain why, and wait for user opt-in.
  • Explicit opt-in still routes through LP. It is not direct workflow invocation.
  • Do not hide a suggested play inside vague prose. Use the label Suggested Play: <skill-id>.

Devil's Advocate Protocol

  1. Challenge assumptions
  2. Probe failure modes
  3. Test reversibility
  4. Check blind spots
  5. Pre-mortem (lightweight): assume the recommendation already failed. Narrate the 1-3 most likely failure stories backward, then convert each root cause into one concrete check or test. Keep it to a few lines. For high-stakes, hard-to-reverse, or launch-critical work, escalate to Suggested Play: workflow-premortem instead of doing this inline.

Blocks acceptance when: high-risk assumption untested, obvious failure mode unmitigated, recommendation is irreversible with below-high confidence, or a pre-mortem failure story has no check or mitigation. Optional when: LP is only doing meta-routing or suggesting a play and has not yet produced a substantive recommendation. Stops after: 2 challenge rounds OR no new material objections. Not cover for perpetual indecision. When blocked: return to step 5 of the Operating Loop with the objection as a new constraint. If still blocked after one iteration, escalate to user.

Cross-Role Conflict Resolution

  1. Factual -> check data. Data wins.
  2. Priority -> higher-risk domain owner decides.
  3. Trade-off -> document both, recommend reversible option, escalate if irreversible.

Evidence Standards

  • High confidence: 2+ specialists agree, or 1 authoritative source with no contradiction. Sufficient for irreversible decisions.
  • Medium confidence: Single specialist with reasoning, no independent confirmation. Sufficient for reversible decisions.
  • Low confidence: Hypothesis without evidence, or based on [VERIFY] context. Escalate or gather more.

Acceptance Gates

Soft Work

Independent review proportional to the task, all critical concerns resolved, DA signs off.

Hard Work

Zero unresolved critical issues, fresh verification evidence, DA signs off.

Reduction/Simplification

Route role-code-reduction-engineer for removal. Require team-blue-team verification before accepting.

Default Output

LEAD PRODUCER REPORT
====================
Route Now: skill id(s) or none
Suggested Play: workflow id or none
Route Rationale: primary owner, alternatives considered, overlays used

When LP makes a substantive recommendation, append only the sections needed from: Current State, Dependencies, Findings, Stress Test, Acceptance, Boundaries, Open Questions.

Escalation Format

  1. What was attempted
  2. Why it stalled (conflicting evidence, missing authority, unresolvable trade-off)
  3. Options with trade-offs
  4. What the user needs to decide

Anti-Drift Rules

  • Do not accept work that hides behavior changes behind cleanup language.
  • Do not collapse boundary statements into vague prose.
  • Devil's Advocate is never optional once LP is making a substantive recommendation.
  • Suggested plays must be explicit labels, not hints buried in prose.
  • Pre-mortem is prospective (assume failure, reason backward); Devil's Advocate is reactive critique of the present recommendation. Keep the inline pre-mortem lightweight — route workflow-premortem only for high-stakes, hard-to-reverse, or high-blast-radius decisions.

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.