Orchestrate Analyze

--- name: orchestrate-analyze description: Orchestrated Delivery O1: triage a feature, spawn conditional Task 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

1. Read E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/rules/guardrails.mdc
2. Read _shared/sdd-artifacts/SESSION.md; load session-state for $Cwd
3. If the relevant gate is not approved: STOP - ask user (pt-BR) - do NOT Write/Shell
4. SDD/develop skills: after ONE step/task, STOP session - handoff only
5. This skill body is English; user-facing prompts may be (pt-BR)

Step -1 - Gate check (report in chat before continuing)

Gate check:
[ ] guardrails.mdc read
[ ] SESSION.md read; session-state loaded
[ ] PIPELINE.md read (required for orchestrate-*)
[ ] User confirmed current action (sim)
-> If any unchecked: STOP

Skill: orchestrate-analyze

Trigger

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)
3. features/NNN-slug/USnn/STORY.md and/or TSnn/STORY.md - BDD + scorecard summary + deps

Step 0 (required): Memory Bank Gate (MEMORY-BANK.md, policy auto) before triage. Resolve bank_root via STORAGE.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). 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 (SKILL §7b).

Orchestrator does not implement application code. Specialists write notes under story ANALYSIS/ / ARCH/ / SEC/ only when spawned.

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

Lazy-load

Process

Step -1b - Caveman Mode (Lite cap)

1. Read E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/sdd/preferences.json (create { "caveman_mode": false, "caveman_level": "full" } if missing).
2. If caveman_mode is false: continue without compression.
3. If true: load E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/caveman/CAVEMAN.md; apply Lite participation cap + prefs caveman_level (Lite skills never escalate); show once: [Caveman] Modo ativo (respostas compactas, level={effective}). Digite caveman off para desativar.
4. Honor caveman on|off|status|lite|full|ultra (and stop caveman / normal mode) during the session.
5. Auto-Clarity + never-compress gates/drafts/paths per CAVEMAN.md.

1. Gate check

Report the Step -1 gate checklist in chat. Load PIPELINE.md (Orchestrated Delivery) and SESSION.md. STOP if any gate unchecked.

2. Resolve storage

Load STORAGE.md. Run resolution with $Workflow = classic. Resolve feature root and bank root:

• repository -> feature $Cwd/features/; bank $Cwd/memory-bank/
• global -> feature <classic.path>/features/; bank <classic.path>/memory-bank/

Path sanitize (required) for any invoke / allocated feature path: normalize (\ -> /, trim trailing /, resolve .). Reject if it contains .., or if the resolved absolute path is not under the feature root above. Ask again in pt-BR for a canonical path - do not Read/Write outside the feature root.

If first run for this repo: ask storage (pt-BR) per STORAGE.md and persist manifest. Confirm target workspace. Do not invent a feature path outside the resolved root.

Repository mode: ensure SDD .gitignore patterns per STORAGE.md (includes /features/; do not add /memory-bank/ — commit bank when product knowledge; never commit secrets) when writing under features/ or memory-bank/ (do not weaken toolkit patterns; never ignore skills/). Global mode: do not edit project .gitignore.

3. Step 0 - Memory Bank Gate

Follow E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/sdd-artifacts/MEMORY-BANK.md (policy default auto). Bank root = resolved bank_root from step 2 - never under features/NNN-slug/.

Before any bank write: confirm (pt-BR) per MEMORY-BANK.md / guardrails (sim / ajustar / cancelar). Healthy bank -> selective read only (no write). Explicit skip / skip-memory-bank -> log and continue (exception only).

Record bank_path + status (fresh | created | refreshed | skipped) for CONTINUITY (steps 6 and 8).

Must not: dump the entire bank into the parent prompt; create bank under features/; write app code during Step 0.

4. Collect description and triage

Ask for (or reuse Prior context): goal, current behavior, constraints, known repos/areas. Use selective memory-bank facts as Prior context - do not re-ask what the bank already states clearly.

Set and record:

Set needs_* flags (needs_api, needs_domain, needs_database, needs_frontend, needs_security, needs_devops) using the canonical table in ROSTER.md (do not fork a second mapping here). Optional NuGet/examples: reference.md.

TE01 - ambiguous flags: ask at most a few high-cost questions (pt-BR). Do not invent architecture in the orchestrator. Prefer false until evidence or user confirms - except auth / secrets / PII / feed-token / supply-chain signals -> ask explicitly or set needs_security=true.

Suggest path (RF01):

5. Trivial shortcut

If trivial: recommend skipping full O1 write:

Escopo trivial. Prefere atalho?

1) /developer  (ou *-developer do stack)
2) Continuar O1 mesmo assim (gravar feature tree)
3) cancelar

Only continue to step 6+ if the user explicitly chooses 2.

6. Allocate NNN-slug and scaffold tree

1. Glob existing NNN under features/*/ only (workspace + global feature root) per STORAGE.md. Next = max + 1. Do not number from root/flat PRD/ or PLAN/.
2. Propose NNN-slug (kebab-case) and full path.
3. Confirm before first Write (pt-BR): “Posso gravar a árvore em {path}? (sim / ajustar / cancelar)” - silence ≠ approval.
4. Create from templates: FEATURE.md, CONTINUITY.md (include Memory-bank path + status from Step 0), story folders USnn/TSnn as needed. Optional subfolders (ANALYSIS/, ARCH/, SEC/, REFINE/) on demand under the story - never at repo root. Do not create PRD/ / PLAN/ yet (O2). Do not create memory-bank/ under the feature path.

7. Spawn Task specialists (conditional, parallel)

SPAWN first: load SPAWN.md; consult capability subagents (native | none). Prefer Task when native; if subagents=none or Task unavailable → fallback in-parent (same specialist notes under ANALYSIS/ / ARCH/ / SEC/, or handoff note in CONTINUITY) — never hard-fail for missing Task. Concurrent Task cap ≤4 per SPAWN.md (wave if more; do not invent a new cap).

Spawn a Task subagent only when subagents=native and ROSTER.md canonical needs_* / brownfield rules say so. Load prompt from skills/_shared/agents/prompts/. When multiple specialists apply, spawn in parallel within the SPAWN ≤4 cap; if more flags apply, batch in waves of ≤4 or ask (pt-BR) to run série.

Model (SUBAGENT-MODEL.md): omit Task model by default (inherit parent / Auto). Ask about a premium slug only for very hard work per that contract; on não / silence, spawn without model. Never pick a costlier model alone.

Parent keeps lean context: synthesis + paths. Specialists must not write app code. Do not call *-developer to implement. qa_checklist is CONTINUITY/STORY only - never spawn a Task for it.

7b. Architecture confirm gate (greenfield / needs_domain)

When nature is greenfield or needs_domain=true and no established in-repo / approved ARCH style:

1. Architect returns an ARCH draft only (four sections per prompts/architect.md; may use templates/features/story/ARCH/architecture-decision.md).
2. Parent presents the draft and asks (pt-BR):

Rascunho ARCH (estilo proposto) em `{story-arch-path}`.

Posso gravar o ARCH aprovado com este estilo?
(sim / ajustar / cancelar)

3. Act on the answer:

4. Until sim: do not write final ARCH; do not invent a silent default style (including vertical-slice). Brownfield with an established style: skip this gate — discover/mirror only (no re-pick).

Canonical details: ROSTER.md architecture confirm gate + reference.md § Architecture confirm gate.

8. Synthesize artifacts

Merge specialist notes + user input into:

1. FEATURE.md - overview, story index, all needs_*, status draft
2. CONTINUITY.md - phase analyze, decisions, flags, open items, Memory-bank path + status (fresh | refreshed | created)
3. STORY.md per US/TS - template structure; BDD Given/When/Then; deps; scorecard summary (rubric from refine-story/reference.md; map /100 -> 1-5 in STORY table)

Use generate-story prompt patterns for drafts. Prefer pt-BR artifact prose; paths/ids English. CONTINUITY references the bank only - do not paste bank body into CONTINUITY.

9. Human backlog approval (RN01)

Present the backlog (feature summary + story table + scorecard highlights). Ask (pt-BR):

Backlog O1 pronto em `{feature-path}`.

Posso marcar como aprovado e seguir para O2?
(sim / ajustar / cancelar)

10. Approve -> CONTINUITY + O2 handoff

On sim:

1. Update FEATURE.md / story statuses to approved as appropriate.
2. Update CONTINUITY.md: phase stays analyze until O2 starts (or set handoff-ready note); Last agent = orchestrate-analyze; keep Memory-bank fields; typed handoff string.
3. Offer O2 (document series vs parallel as O2 choice - do not implement O2 here):

/orchestrate-deliver - <full-feature-path>

Example:

/orchestrate-deliver - features/004-nuget-extract/

Remind (pt-BR): O2 will ask série vs paralelo for per-story PRD/PLAN.

11. Context pressure (TE02 / RNF02)

Honor E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/rules/context-management.mdc thresholds (checkpoint / hard stop). When pressure is high:

1. Persist latest CONTINUITY.md (estado atual short per CONTINUITY template, decisões, pendências, exact next /…).
2. Offer session handoff - same phase, resume with feature path:

/orchestrate-analyze - <full-feature-path>

Do not paste full specialist dumps into the parent chat.

Must not

• Skip Step 0 Memory Bank Gate (unless explicit user skip-memory-bank)
• Create or place memory-bank/ under features/NNN-slug/
• Duplicate memory-bank body into CONTINUITY (path + status only)
• Dump entire memory-bank into the parent orchestrator context
• Write application/production code or tests (*.cs, *.tsx, *.ts, *.js, *.vue, *.py, migrations, etc.)
• Call *-developer / developer to implement code (suggesting the trivial shortcut is allowed)
• Skip human backlog approval or treat silence as sim
• Skip the architecture confirm gate on greenfield / needs_domain (no established style), or write final ARCH / pick a silent style default before operator sim
• Write PRD/PLAN (O2 owns that via sdd-spec / sdd-plan contracts)
• Create external work-item tracker or org-only compliance content
• Create ~40 agent files or expand the roster beyond ROSTER.md
• Modify toolkit .gitignore as part of porting this skill into the toolkit repo; at runtime follow STORAGE.md only for consumer repo SDD patterns
• Change the sdd-develop one-step-per-session contract
• Create REFINE/ / ANALYSIS/ / ARCH/ / SEC/ / PRD/ / PLAN/ at repo root
• Resolve feature paths outside $Cwd/features/ or <classic.path>/features/, or accept .. segments
• Pass Task model without SUBAGENT-MODEL.md gate + user sim (or user-named slug); ask model on routine spawns
• Hard-fail when subagents is none or Task is unavailable (use fallback in-parent / handoff per SPAWN.md)
• Exceed orchestrate ≤4 concurrent Tasks without user-approved wave/série (SPAWN.md)
• Paste guideline packs into Task child prompts

Handoff

Canonical O2 handoff (exact pattern)

/orchestrate-deliver - <full-feature-path>