Skip to content

tibursocampos/agent-dev-toolkit / core/skills/orchestrate-analyze

orchestrate-analyze

Unverified

Orchestrated Delivery O1: triage a feature, spawn conditional Agent specialists, write FEATURE.md + CONTINUITY + US/TS under features/NNN-slug/. No app code. Use when invoking /orchestrate-analyze.

STOP - Read before ANY tool call

Read _shared/skill-boilerplate.md first. It holds the guardrails read, the session gate rule, the user sim rule, the one-step rule, the gate check, and the user chat language rule. The steps below are skill-specific only.


Skill: orchestrate-analyze

Trigger

Negative signal: applies as defined in _shared/skill-boilerplate.md; the skill-specific signals are listed below.

Invoke when the user asks for: /orchestrate-analyze, orchestrate analyze, /orchestrate-analyze, or Orchestrated Delivery analysis for a complex / multi-story / brownfield feature.

Optional: pasted feature description, existing notes path, or prior refine output.

Outcome

Under the resolved classic feature root (STORAGE.md, $Workflow = classic):

  1. features/NNN-slug/FEATURE.md - triage, scope, nature, complexity, needs_*
  2. features/NNN-slug/CONTINUITY.md - phase, decisions, typed handoff, Memory-bank path + status (fresh | refreshed | created; refreshed after ARCH sim / point-promote)
  3. features/NNN-slug/USnn/STORY.md and/or TSnn/STORY.md - BDD + scorecard summary + deps. Create these only at step 9, and only when FEATURE has no open question.

Step 0 (required): Memory Bank Gate (MEMORY-BANK.md, policy auto) before triage. Resolve bank_root via STORAGE.md + STORAGE-RESOLUTION.md ($Cwd/memory-bank/ or <classic.path>/memory-bank/) - never under features/NNN-slug/.

Human gate: backlog must be explicitly approved (sim / ajustar / cancelar) before O2. Silence is not approval (RN01). Do not mark approved if required specialist folders are missing or cited non-feature .md was not promoted (pointer-only = fail O1). When greenfield / needs_domain without an established style, an architecture confirm gate (ARCH draft → sim → ARCH approved) runs after the architect pass and before treating style as selected (references/arch-confirm.md). Brownfield skips style re-pick only — still write mirror ARCH.

Orchestrator does not implement application code. At steps 8 and 8b, hold specialist notes in the parent. Create story folders and ANALYSIS/ / ARCH/ / SEC/ only at step 9, and only when FEATURE has no open question. REFINE/ remains on demand. Do not route those notes to CONTINUITY as a substitute.

Does not write PRD/PLAN (that is O2 via sdd-spec / sdd-plan contracts). Does not call trackers.

Mental map (ids unchanged): O1 ≈ explore (FEATURE, stories, specialists). O2 ≈ FEATURE+PRD+CHANGE; O3 ≈ apply. See CHANGE-CONTRACT.md.

Lazy-load (only when needed)

WhenPath (after syncing the active adapter)
Command playbook (step discovery after gates)references/command.md
Caveman Mode (if active){{TOOLKIT_ROOT}}/skills/_shared/caveman/CAVEMAN.md - Lite cap
Pipeline Orchestrated Delivery, confirm, paths{{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/PIPELINE.md
Invocation contexts (direct vs orchestrated, IC-DIRECT-ORCHESTRATED){{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/INVOCATION-CONTEXTS.md
Evidence decisions, relevant sources, decoys, gaps, and conflicts (SR-NO-FULL-DUMP){{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/SELECTIVE-RETRIEVAL.md
Material behavior or decision carried into FEATURE/STORY (Steps 5 and 9 only){{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/SEMANTIC-SUFFICIENCY.md
Storage, manifest, feature tree{{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/STORAGE.md
Dynamic manifest / repository-global root resolution (only when resolving storage){{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/STORAGE-RESOLUTION.md
Step 1 triage — intent classification (before Step 0 when not full O1)references/intent-classification.md
Step 0 Memory Bank Gate{{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/MEMORY-BANK.md
Memory-bank create/refresh{{TOOLKIT_ROOT}}/skills/memory-bank-init/SKILL.md
Roster, needs_*, triage table{{TOOLKIT_ROOT}}/skills/_shared/agents/ROSTER.md
Spawn native vs fallback (capability subagents){{TOOLKIT_ROOT}}/skills/_shared/agents/SPAWN.md
Agent subagent model (default omit; rare premium gate){{TOOLKIT_ROOT}}/skills/_shared/agents/SUBAGENT-MODEL.md
Stack routing (implement later){{TOOLKIT_ROOT}}/skills/_shared/agents/ROUTING.md
Templates{{TOOLKIT_ROOT}}/skills/_shared/templates/features/{FEATURE,CONTINUITY,TREE}.md, .../story/STORY.md
Specialist prompts{{TOOLKIT_ROOT}}/skills/_shared/agents/prompts/{repo_analyst,architect,security,database,impact,risk,generate-story}.md — one file per spawned specialist
Reference index (routing only)reference.md
Process step detail (lazy)references/<section>.md
Scorecard rubric (reuse){{TOOLKIT_ROOT}}/skills/refine-story/references/scorecard-rubric.md
Scorecard template (optional shape){{TOOLKIT_ROOT}}/skills/refine-story/references/scorecard-template.md
Story sizing (synthesis / merge policy){{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/story-sizing.md
Persona / Product intent (US only){{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/persona-context.md
Promotion / anti-task-shatter gate (synthesis){{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/anti-task-shatter.md
Cap / altitude rationale (synthesis){{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/feature-altitude.md
Step 8c feature research (before story folders){{TOOLKIT_ROOT}}/skills/refine-story/references/feature-research.md
Live stage table (step 8c chat){{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/LIVE-STAGE-TABLE.md

Never by default: do not preload references/command.md before Step -1 gates; do not preload all specialist prompts, all templates, or ROSTER+SPAWN+SUBAGENT-MODEL+ROUTING+MEMORY-BANK together. Do not dump guideline packs into Agent child prompts. Do not preload persona-context.md except when synthesizing User Story Product intent. Do not preload anti-task-shatter.md / feature-altitude.md until synthesis / product artifact gates.

Progressive load: contracts first (PIPELINE.md + STORAGE.md); after gates load references/command.md for step discovery; fan-out only on trigger (MEMORY-BANK.md at Step 0, ROSTER.md when setting needs_*, SPAWN.md before Agent, one prompt per spawn, one references/<section>.md per Process step — never full reference.md when a section file exists (SKILL-REFERENCE-RETRIEVAL.md).

Reference routing

SituationPath
Command playbook (step discovery)references/command.md
Gate / Orchestrated Delivery / promotePIPELINE.md
Feature root / bank root / portable pathsSTORAGE.md
Step 0 Memory BankMEMORY-BANK.md
Set needs_* / brownfield spawn rulesROSTER.md
Before any Agent / fallbackSPAWN.md
Premium Agent model gate (rare)SUBAGENT-MODEL.md
Implement-later stack hintROUTING.md
Step 1 intent classification (triage entry)references/intent-classification.md
Step 0 Memory Bank detailreferences/memory-bank-gate.md
Triage / collect / trivial / NuGetreferences/triage.md
Spawn / specialist mapreferences/spawn-map.md
ARCH confirm gatereferences/arch-confirm.md
Step 8c feature research (before story folders){{TOOLKIT_ROOT}}/skills/refine-story/references/feature-research.md
Live stage table (step 8c chat){{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/LIVE-STAGE-TABLE.md
Scaffold / synthesize / CONTINUITY / product artifact gatesreferences/story-synthesis.md
Boundaries / handoff stringsreferences/boundaries-handoff.md
Caveman / storage / approval / contextreferences/process-common.md
Must not (full)references/must-not.md
Scorecard /100 → STORY{{TOOLKIT_ROOT}}/skills/refine-story/references/scorecard-rubric.md
Story sizing / merge policy{{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/story-sizing.md
Product intent (US Who/Job/Outcome){{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/persona-context.md
Promotion anti-task-shatter{{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/anti-task-shatter.md
Cap / feature altitude{{TOOLKIT_ROOT}}/skills/_shared/backlog-item-types/feature-altitude.md

Process

After gates: Read references/command.md for ordered step discovery (do not dump this Process section into child prompts). Then load references/<section>.md for the current step only — not full reference.md. Do not skip gates.

Step -1b - Caveman Mode (Lite cap)

Apply Lite caveman prefs when active. Read references/process-common.md § Process — Caveman (Lite cap).

1. Gate check

Report the Step -1 gate checklist in chat. Load PIPELINE.md (Orchestrated Delivery) and SESSION.md. STOP if any gate unchecked. Stay on the current branch. Do not show the branch and do not ask for a branch switch, including when the branch is main, master, or develop.

1b. Resolve invocation context

Load INVOCATION-CONTEXTS.md. This skill defaults to orchestrated (IC-DIRECT-ORCHESTRATED). Apply orchestrated observable rules; pass invocation_context: orchestrated into child handoffs (path cite only — do not dump the contract body).

2. Resolve storage

Load STORAGE.md ($Workflow = classic); resolve feature + bank roots; path sanitize; gitignore per STORAGE. Read references/process-common.md § Process — Resolve storage.

3. Intent classification (Step 1 triage entry)

Load references/intent-classification.md. Classify input (Existing Feature | New Feature | Product Initiative | Problem/Need | Idea); map to Classic SDD vs Backlog Refine vs full O1. On early handoff (/refine-story, /sdd-spec, /developer), confirm in the user chat language and STOP — skip Step 0 unless operator chooses to continue O1 anyway. Record intent when continuing full O1.

4. Step 0 - Memory Bank Gate

Follow MEMORY-BANK.md (policy default auto). Bank root = resolved bank_root — never under features/NNN-slug/. Confirm before bank write; healthy → selective read only. Record path + status for CONTINUITY; after ARCH sim point-promote architecture.md and set refreshed. Read references/memory-bank-gate.md.

5. Collect description and triage

Ask/reuse Prior context; promote cited non-feature .md per PIPELINE.md. Set complexity/nature/scope + needs_* via ROSTER.md (TE01 / RF01). Read references/triage.md (after intent from step 3).

For each analysis question that can affect scope, story boundaries, architecture, security, or the O2 handoff, apply SELECTIVE-RETRIEVAL.md § Evidence decisions. Record the question/objective, relevant source-of-truth paths, rejected decoys with the scope reason, remaining gaps or conflicts, the decision and its source, and the next permitted gate in existing FEATURE / CONTINUITY sections. Keep this as traceable summary, not a second source of truth. If a gap or conflict can change scope or acceptance, keep it open and do not present the backlog as ready. Label operator reports and comparative material as such; do not promote them to independent implementation or runtime proof. For external evidence intended to be reachable during O3, record the exact external:<alias>/<relative-path> reference in the existing evidence/source note, with no root or absolute path (STORAGE.md § External evidence references for O3). This captures a reference only; O1 does not authorize its alias.

For brownfield work, apply {{TOOLKIT_ROOT}}/skills/_shared/agents/BROWNFIELD-ANALYSIS.md proportionally to the affected scope. O1 synthesis must check that applicable conventions and impacts have repository evidence, confidence, reuse/affected areas, and material gaps recorded in the existing analysis/architecture artifacts. Folder or artifact presence alone is not evidence of sufficient analysis. Keep technical pattern investigation (targeted repository inspection) distinct from product questions: only unresolved decisions about behavior, scope, acceptance, security, ownership, or validation become operator questions. If a material technical evidence gap remains after feasible inspection, record it and do not present dependent stories/backlog as ready; preserve O1 ownership of FEATURE/CONTINUITY and specialist-owned findings.

When that investigation will shape a material story boundary or downstream acceptance, also apply SEMANTIC-SUFFICIENCY.md to the specific content consumed. Resolve the current decision and compare its actual normative destination before carrying a historical source forward; record the disposition in existing FEATURE fields rather than a second decision list.

6. Trivial shortcut

If trivial: offer shortcut (read references/triage.md § Process — Trivial shortcut). Continue O1 only if user chooses 2.

7. Allocate NNN-slug and scaffold tree

Glob next NNN; confirm portable path; scaffold FEATURE.md and CONTINUITY.md only. Do not create story folders. Read references/story-synthesis.md.

8. Spawn Agent specialists (conditional, parallel)

SPAWN first; spawn per ROSTER when native; fallback in-parent write; cap ≤4; omit Agent model by default. Read references/spawn-map.md.

8b. Architecture confirm gate (greenfield / needs_domain)

ARCH draft → operator sim / ajustar / cancelar → approved + point-promote. Brownfield: skip style re-pick only; still write mirror ARCH. Read references/arch-confirm.md.

8c. Feature research before story folders

Load {{TOOLKIT_ROOT}}/skills/refine-story/references/feature-research.md. Update only FEATURE.md and CONTINUITY.md. If any open question remains, stop with open_question. Do not run step 9.

You must render, in the user chat language, the 8c stage table (stage, status, weight, what_happened; the six O1 rows, including the second critique round) and the cumulative checkpoint defined in {{TOOLKIT_ROOT}}/skills/_shared/sdd-artifacts/LIVE-STAGE-TABLE.md. Labels, state keys and ASCII glyphs come only from that file. You must render both at each named display point: the start of the stage, before each gate, after each gate answer, on each row state change, on a blocker (open_question, with the blocked row blocked), and at the close of 8c. Completeness rule: always render the full table with every row; never summarize rows, show only the changed row, or omit completed rows. Both are chat-only and never persisted in FEATURE.md or CONTINUITY.md.

9. Synthesize artifacts + product artifact quality gates

Run only when FEATURE has no open question and any required brownfield analysis is sufficient for the dependent story scope. Load story-sizing.md; apply merge policy (merge file/layer fragments; split when >~8 refine steps or independent consumers). Lazy-load anti-task-shatter.md (and feature-altitude.md when cap/altitude unclear). For each material behavior, apply SEMANTIC-SUFFICIENCY.md before promotion: compare the current decision with its actual normative destination and the candidate story acceptance text. Reuse existing macro-acceptance, Decisions, Open questions, and story-index fields to show whether the behavior is covered, preserved unchanged, explicitly excluded, or still a question; valid links or IDs alone do not pass this check. Create story folders and merge into FEATURE / CONTINUITY / STORY (scorecard via {{TOOLKIT_ROOT}}/skills/refine-story/references/scorecard-rubric.md). Each new STORY carries type, objective, and the FEATURE acceptance ids it covers. FEATURE.md story table must include Rationale and Product intent per row (Who/Job/Outcome or n/a; lazy-load persona-context.md for User Stories only). Before Step 10: run product artifact quality gates in references/story-synthesis.md — FEATURE depth (TE01 / Problem+Goals+Non-goals), promotion anti-task-shatter (TE02 — no US/TS for verb+file/class/script or layer-only), cap ≤4 US/TS unless explicit split rationale (RN03), and the deterministic evidence validation for every required brownfield ANALYSIS/ document. Any fail → keep draft; return invalid analysis to its owner for correction, list fields/rules; do not present human approval. Read references/story-synthesis.md.

10. Human backlog approval (RN01)

Before presenting the backlog, verify the evidence summary required in Step 5 is persisted in the existing FEATURE / CONTINUITY fields and identifies the next gate. Product artifact quality gates must have passed. For each required brownfield ANALYSIS/, verify the Step 9 validator returned PASS; folder presence alone is insufficient. Required folders + promote first; present backlog; sim / ajustar / cancelar (silence ≠ approval). O2 retains its independent brownfield validation gate. Read references/process-common.md § Process — Backlog approval + O2 handoff and references/arch-confirm.md § Approval gate copy.

11. Approve -> CONTINUITY + O2 handoff

On sim: approve statuses; typed O2 handoff. Read references/process-common.md § Process — Backlog approval + O2 handoff and references/boundaries-handoff.md § Canonical handoff strings.

12. Durable progress checkpoint

Checkpoint durable workflow progress at the workflow-defined boundaries; context usage does not add a stop gate.

Must not

Enforce the full list in references/must-not.md. Critical always-on: no app code; no PRD/PLAN; no skip Step 0 / backlog sim / ARCH confirm; no missing ANALYSIS|ARCH|SEC when flags require them; no hard-fail when Agent unavailable (fallback in-parent write); portable paths only; do not ignore IC-DIRECT-ORCHESTRATED (INVOCATION-CONTEXTS.md); do not present human backlog approval when FEATURE depth / promotion / cap gates fail; do not promote task-shaped or layer-only titles to US/TS.

Handoff

SituationNext
Backlog approved/orchestrate-deliver - <portable-feature-path>
Handoff mid-O1/orchestrate-analyze - <portable-feature-path>
Trivial after triage/developer or stack *-developer
Single clear story, skip O2 multi/sdd-spec (Classic SDD) after STORY exists
Informal single item only/refine-story (Backlog Refine)

Canonical O2 handoff (exact pattern)

/orchestrate-deliver - <portable-feature-path>
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 467kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 319kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 319kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 319kUpdated Claude CodeCodexVerifiedInstall