orchestrator

Orchestrator

The central control skill for agent-driven projects. It plans, governs, audits, delegates, and re-validates. It never executes complex work directly when a specialized skill exists.

All questions and confirmations directed at the user must be in Portuguese (pt-BR). Internal reasoning and documentation are in English.

When to Use

• Starting a new project or repository.
• Resuming an existing project with unclear state.
• Planning a feature, Epic, or release.
• Coordinating implementation of a SPEC SDD.
• Preparing a PR after implementation.

When NOT to Use

• Do not use when the task is a single, well-scoped code change — use /tdd-spec directly.
• Do not use when only a code review is needed — use /code-review-and-quality.
• Do not use when only a bug fix is needed — use /diagnose.

State File

The Orchestrator must read references/ESTADO_ORQUESTRATOR.md at the start of every session and write to it after every phase. This state file persists the DAG, task status, and decisions across sessions. See references/ESTADO_ORQUESTRATOR.md.

Phase -1 — Framework Update

Run this at the start of every Orchestrator session, before project preconditions.

1. Find where the skills were installed from. For each loaded skill, resolve the real path of the link and locate the catalog clone that contains README.md and SKILL.md.
2. In the found clone, read origin remote, current branch, and local installed commit.
3. Check the framework remote with git fetch origin --quiet. Never pull, merge, or reset the framework clone.
4. Compare local commit with origin/ or the equivalent remote reference.
5. If there are new commits, report immediately:

Framework update available
- Framework: afonsoft/skills
- Installed: 
- Available: 
- Changes: 
- Action: reinstall the catalog with `npx skills add afonsoft/skills`

1. If new commits are available, guide the user to reinstall skills with npx skills add afonsoft/skills.
2. After reinstall, confirm orchestrator and create-agent-harness point to the new revision and report the result.
3. If no changes, log Framework up-to-date () without stopping the flow.
4. If the clone, remote, or network cannot be located, log Unable to check framework updates and continue only if local skills are available. Do not reinstall without confirming a new revision.

When a new revision is confirmed, the user must reinstall the skills. That is part of the Orchestrator contract.

Phase 0 — Governance Preconditions

Before creating files or delegating work:

1. Verify Git is initialized.
2. Verify a valid GitHub remote exists, preferably origin.
3. Verify repository access with gh repo view or equivalent.

If the environment is empty, has no Git, or has no GitHub remote, stop the flow and guide the user to:

1. Create the repository on GitHub;
2. Initialize the local repository;
3. Configure the origin remote;
4. Make the first commit and push;
5. Return to the Orchestrator.

Never silently replace GitHub with a local tracker. GitHub is the source of traceability, Issues, review, and history for this framework.

Phase 1 — Documentation Provisioning

1. Invoke /create-agent-harness to generate CLAUDE.md, AGENTS.md (thin reference), .claude/ (settings, rules, agents, memory, context), docs/ (technologies, architecture, decisions), and .specs/.
2. Invoke /grill-me-with-spec to consolidate domain language and architectural decisions, producing the SPEC SDD in .specs/SPEC-{YYYYMMDD}-{feature}.md before any implementation.
3. In an empty repository, invoke /scaffold-mvp after domain alignment.
4. Review and persist documentation and the approved SPEC before starting implementation.

Documentation is not optional: the Orchestrator must leave a state another agent can continue.

Special Case — New Project with Only a PRD in the Folder

When the repository starts from a folder containing only a PRD (no code):

1. Ensure GitHub repository is initialized with origin configured (Phase 0).
2. Create and check out a develop branch from the default branch.
3. Invoke /grill-me-with-spec to turn the PRD into one or more SPEC SDDs in .specs/SPEC-{YYYYMMDD}-{slug}.md, one per Epic or well-delimited area.
4. Review and approve the SPECs; update Status to Approved on each one.
5. Based on approved SPECs, open Issues on GitHub using /create-issues (one per Epic, or a master Issue with Epics listed).
6. Use /create-issues to slice each Epic into atomic Issues (vertical, traceable, with acceptance criteria), recording the mapping .specs/SPEC-*.md → Issue.
7. Proceed to Phase 4 using the sequential queue described below.

Phase 2 — Audit

Audit the structure produced by create-agent-harness:

[ ] Git initialized
[ ] GitHub remote configured and accessible
[ ] CLAUDE.md (single source of truth) and AGENTS.md (thin reference)
[ ] .claude/settings.json (permissions, hooks, env)
[ ] .claude/rules/global-rules.md and stack-scoped rules/
[ ] .claude/agents/ (review.md, plan.md, test.md)
[ ] .claude/memory/ and .claude/MEMORY.md
[ ] .claude/CONTEXT.md, .claude/RULES.md, .claude/TOOLS.md, .claude/WORKFLOWS.md
[ ] .claude/README.md (harness infrastructure)
[ ] .specs/ for SPEC SDD when features are in flight
[ ] docs/agents/ when domain tracker and labels exist
[ ] docs/adr/ when relevant architectural decisions exist
[ ] Skills installed in the chosen environment

Classify gaps as P1 (security/types), P2 (architecture), P3 (performance), or P4 (hygiene/documentation). To analyze and address gaps, invoke /improve-codebase-architecture.

Phase 3 — GitHub Fragmentation

Approved gaps must be turned into Issues by /create-issues. GitHub is the persistent source of scope, acceptance criteria, dependencies, and status; references/ESTADO_ORQUESTRATOR.md is only the operational view of the DAG.

1. Pass the gaps, roadmap, and approved documentation to /create-issues.
2. Present the decomposition for approval when HITL decision is needed.
3. Publish Issues in dependency order, using real IDs in Blocked by.
4. Record the mapping Task -> GitHub Issue -> branch/worktree.
5. Never create a DAG only in memory or only in a local file when the task can be tracked on GitHub.

Phase 4 — Execution Loop

flowchart TB
    subgraph Phase1["Phase 1 - Plan"]
        P1_H[/create-agent-harness/]
        P1_S[/grill-me-with-spec/]
        P1_M[/scaffold-mvp/]
    end

    subgraph Phase2["Phase 2 - Audit"]
        P2_A["Audit gaps"]
        P2_F[/improve-codebase-architecture/]
    end

    subgraph Phase3["Phase 3 - Issues"]
        P3_I[/create-issues/]
    end

    subgraph Slice["Per-Slice Loop"]
        S_R["Read SPEC + Issue"]
        S_T[/tdd-spec/]
        S_C[/code-review-and-quality/]
        S_D[/diagnose/]
        S_V["Verify build / test / lint"]
        S_G[/grill-me-with-spec/]
        S_CM["Commit"]
    end

    subgraph Gate["Phase 5 - QA"]
        G_Q[/qa-analyst/]
        G_R[/code-review-and-quality/]
        G_M[/create-readme/]
        G_P["PR / Merge"]
    end

    P1_H --> P1_S
    P1_S --> P1_M
    P1_M --> P2_A
    P2_A -->|P2 gap| P2_F
    P2_F --> P2_A
    P2_A --> P3_I
    P3_I --> S_R
    S_R --> S_T
    S_T --> S_C
    S_C --> S_V
    S_V -->|green| S_CM
    S_T -->|bug| S_D
    S_D --> S_T
    S_C -->|ambiguous| S_G
    S_G --> S_R
    S_V -->|fail| S_D
    S_CM -->|next slice| S_R
    S_CM -->|Epic done| G_Q
    G_Q -->|approved| G_R
    G_Q -->|fail| S_T
    G_R -->|approved| G_M
    G_R -->|fail| S_T
    G_M --> G_P

The Orchestrator runs sliced Issues in a continuous loop until all SPEC implementations are complete. The focus is small vertical slices, one at a time, with constant re-validation.

General Rules

• Independent slices may run in parallel in isolated worktrees; slices that change schema, authentication, public APIs, or data require human confirmation.
• Before each slice, the agent must read the approved .specs/SPEC-{YYYYMMDD}-{slug}.md and the corresponding Issue.
• After each slice, re-validate: build, tests, lint, type check.
• Do not move to the next slice while the current one is not green.

Per-Slice Cycle

1. READ         → Approved SPEC + GitHub Issue
2. TDD          → /tdd-spec (red-green-refactor) using acceptance criteria
3. CODE REVIEW  → /code-review-and-quality on the slice diff
4. ARCH         → /improve-codebase-architecture if architecture degrades
5. DIAGNOSE     → /diagnose if a bug or mysterious failure appears
6. CLARIFY      → /grill-me-with-spec if the SPEC is ambiguous
7. VERIFY       → build, tests, lint pass
8. COMMIT       → Conventional commit, reference the Issue
9. LOOP         → Next slice in the queue

Skill Delegation by Situation

Sequential Queue for Epics Sliced from a PRD

When Issues come from the special case "new project with only a PRD" (Phase 1), execution is not parallel: dispatch one agent at a time, in Issue dependency order.

1. For the current Epic, process its sliced Issues one by one:
   • develop with /tdd-spec;
   • QA (Phase 5);
   • commit;
   • next Issue in the queue. Repeat until all Issues of the Epic are exhausted.
2. When the Epic is complete, invoke /qa-analyst, then /code-review-and-quality for the accumulated diff, then /drawio-architecture to refresh the system diagram, then /create-readme to reflect what was delivered.
3. Epic exhausted → open a PR from the working branch to develop.
   • Green PR (CI/tests pass) → merge into develop.
   • Failed PR → fix with /diagnose, re-run verification, then merge.
4. After the merge, return to the develop branch and advance to the next Epic in the queue, repeating the loop until all PRD Epics are finished.
5. When all Epics are complete, open the final merge from develop to main.

Phase 5 — Verification and QA

After each slice and at the end of each Epic/DAG:

1. Run proportional verifications: tests, lint, type check, build.
2. If it fails, invoke /diagnose before continuing.
3. When the DAG is complete, invoke /qa-analyst without exception of tier. QA must confront requirements, Issues, implementation, tests, error scenarios, and out-of-scope changes. Failures reopen Issues or create new tasks.
4. After QA approval, invoke /code-review-and-quality for a final review of the accumulated Epic diff (or set of slices). Quality failures reopen Issues or create new tasks.
5. After review approval, invoke /drawio-architecture to update or create the system architecture diagram so documentation reflects the delivered structure.
6. After the architecture diagram is consistent, invoke /create-readme to update README.md with the delivered features, stack, and instructions.
7. Only after that can delivery by PR occur. If no Git/PR flow skill is installed, describe the steps and ask for human confirmation; never invoke a nonexistent skill.

At the end of the project or release, ensure README.md reflects the current system state.

Skill Call Reference

Decision Tree

1. Does the SPEC exist and is Approved?
   • No → /grill-me-with-spec.
2. Is there a P2 architecture gap?
   • Yes → /improve-codebase-architecture.
3. Is the work tracked on GitHub?
   • No → /create-issues.
4. Did a test fail or build break?
   • Yes → /diagnose.
5. Is the code written but not reviewed?
   • Yes → /code-review-and-quality.
6. Is the Epic done and tests green?
   • Yes → /qa-analyst → /code-review-and-quality → /create-readme → PR.

References

• references/orchestrator-delegation-protocol.md — autonomy matrix, risk tiers, and delegation protocols.
• references/ESTADO_ORQUESTRATOR.md — operational state file for the session DAG.
• /create-agent-harness — for generating the project harness
• /grill-me-with-spec — for authoring the SPEC SDD
• /scaffold-mvp — for bootstrapping a new project
• /create-issues — for turning work into GitHub Issues
• /improve-codebase-architecture — for analyzing and fixing architecture gaps
• /tdd-spec — for test-driven implementation from the SPEC
• /code-review-and-quality — for reviewing diffs
• /diagnose — for debugging regressions and bugs
• /qa-analyst — for the mandatory QA gate
• /create-readme — for keeping README in sync