Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsBlogPro
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Authors
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges
  • Chrome Extension
  • Skill Manager

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

ProTermsPrivacyRefunds
Back to skills

Orchestrate Develop

ASecurity

Orchestrated Delivery O3: one Task subagent per PLAN step (sdd-develop contract); parent never writes app code. Updates CONTINUITY; handoff to code-review. Use when invoking /orchestrate-develop.

6 stars
0 votes
0 copies
0 views
Added 9/25/2026
ai-agentsshellrailscode-reviewgit

Security Analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned 9/25/2026

$npx -y skills add tibursocampos/agent-dev-toolkit --skill orchestrate-develop --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Orchestrate Develop?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Orchestrate Develop
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/tibursocampos-orchestrate-develop-531d43f0/badge)](https://www.skillsdirectory.com/skills/tibursocampos-orchestrate-develop-531d43f0)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
Files
SKILL.md
---
name: orchestrate-develop
description: Orchestrated Delivery O3: one Task subagent per PLAN step (sdd-develop contract); parent never writes app code. Updates CONTINUITY; handoff to code-review. Use when invoking /orchestrate-develop.
---

## STOP - Read before ANY tool call

1. Read `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/.grok/rules/guardrails.md`
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-develop

## Trigger

Invoke when the user asks for: `/orchestrate-develop`, `orchestrate develop`, `/orchestrate-develop`, or Orchestrated Delivery O3 after O2 handoff.

Required: full feature path **or** a specific `PLAN/PLAN_NNN_*.md` path under a story.

## Outcome

1. Pending PLAN step(s) executed **only** via Task children that follow the **`sdd-develop` contract** (one PLAN step per child / session)
2. Feature `CONTINUITY.md` updated (phase `develop`, progress, typed next invoke, **Memory-bank** path + status)
3. Handoff to `code-review` (`- single` or `- multi-angle`; skill asks if omitted) and/or next step / next story

**Step 0 (required):** Memory Bank Gate (`MEMORY-BANK.md`, policy `auto`) **before** building the step queue / spawning children. Resolve `bank_root` via `STORAGE.md`. CONTINUITY stays the feature phase/handoff source.

**Step N (after code changes):** when at least one develop child succeeded with app file changes, run `memory-bank-init` mode **`refresh-light`** at `bank_root` before final handoff (`MEMORY-BANK.md` Step N). Confirm (pt-BR) first.

**Parent orchestrator never** writes application code, never marks multiple PLAN steps done in one child, and never bypasses `sdd-develop` gates (`step_confirmed`, tests before complete).

**Alternative (always valid):** user runs manual `/sdd-develop - <full-plan-path> - Step N` without this skill (RF05 / CA5). Manual Classic SDD does **not** require memory-bank (CA7).

## Lazy-load

| When | Path |
|------|------|
| Caveman Mode (if active) | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/caveman/CAVEMAN.md` - **Full cap** |
| Pipeline Orchestrated Delivery, paths | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/sdd-artifacts/PIPELINE.md` |
| Storage | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/sdd-artifacts/STORAGE.md` |
| Step 0 Memory Bank Gate | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/sdd-artifacts/MEMORY-BANK.md` |
| Memory-bank create/refresh | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/memory-bank-init/SKILL.md` |
| Develop contract (source of truth) | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/sdd-develop/SKILL.md` + `reference.md` |
| SESSION gates | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/sdd-artifacts/SESSION.md` |
| CONTINUITY template | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/templates/features/CONTINUITY.md` |
| Anti-bypass, parallelism, handoffs | `skills/orchestrate-develop/reference.md` |
| Task subagent model (default omit; rare premium gate) | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/_shared/agents/SUBAGENT-MODEL.md` |
| Code review (ask mode) | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/skills/code-review/SKILL.md` |
| Context pressure | `E:/Source/Repos/agent-dev-toolkit/scripts/validation/fixtures/install-root/rules/context-management.mdc` |

## Process

### Step -1b - Caveman Mode (Full 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 **Full** 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. Ask user **sim** before spawning the first develop child.

### 2. Resolve feature / PLAN set

Load `STORAGE.md` (`$Workflow = classic`). Resolve feature root and `bank_root` (repository vs global per `STORAGE.md` / `MEMORY-BANK.md`).

**Path sanitize (required):** normalize invoke paths; reject `..` and any resolved path outside `$Cwd/features/` (repository) or `<classic.path>/features/` (global). For a single PLAN path, it must remain under that features root. Ask again in pt-BR if invalid.

Repository mode: ensure SDD `.gitignore` per `STORAGE.md` before any bank write. **Global:** do not edit `.gitignore`.

| Invoke | Action |
|--------|--------|
| Feature path | Glob `**/PLAN/PLAN_*.md` under that feature; build story/PLAN queue |
| Single PLAN path | Work that PLAN only; still update feature `CONTINUITY.md` if present |
| Missing PLAN | **STOP** - suggest O2 or Classic SDD |

```text
Não encontrei PLAN sob `{path}`.

1) /orchestrate-deliver - <full-feature-path>
2) /sdd-plan - <full-prd-path>
3) cancelar
```

`Read` `FEATURE.md` + `CONTINUITY.md` when under a feature. Prefer O2-approved stories; if PLAN exists but approval unclear, ask once (pt-BR) before spawning.

### 3. Step 0 - Memory Bank Gate

Follow `MEMORY-BANK.md` (policy default **`auto`**). Bank root = resolved `bank_root` - **not** under `features/`.

Confirm (pt-BR) before create/refresh; healthy -> selective read only. Update CONTINUITY Memory-bank fields when create/refresh runs. Pass **`bank_path`** into every develop child prompt as read-only Prior context (selective files - never dump).

### 4. Build step queue (deps)

For each PLAN:

1. Parse pending steps (`⏳` / Status Pendente / unchecked).
2. Respect **Deps:** only enqueue a step when dependency steps are **Concluídos** / **Completed**.
3. Default order: one story at a time (finish story A before story B) unless user asks otherwise **and** stories are independent.

Present queue summary (pt-BR): story, PLAN path, next step id/title, deps. Confirm:

```text
Fila O3: próximo = `{plan-path}` Step {N} - {title}.

Posso spawnar o subagente (contrato sdd-develop)?
(sim / ajustar / cancelar)
```

Silence ≠ approval (RN01).

### 5. Spawn exactly one step child (CA5)

**Hard rule:** one Task = one PLAN step = full `sdd-develop` contract.

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

Child must:

1. Load and follow `sdd-develop/SKILL.md` (gates, validate step, git branch, implement, tests, update PLAN, report)
2. Receive **only** that PLAN path + step number + lean Prior context paths (PRD, STORY, CONTINUITY, FEATURE, **`memory-bank/` path**) - not full guideline dumps or full bank body
3. Use **PLAN-scoped SESSION** per `SESSION.md` (`plan-{planHash}.json`, or `plan-{planHash}-step-{N}.json` when this spawn is parallel on the same PLAN)
4. Return: `{ planPath, step, status, files[], testsSummary, nextStep?, blockedReason? }`
5. **STOP** after that step - must not start Step N+1 in the same child

**Parent must not:**

- Edit `*.cs` / app sources / tests itself
- “Help finish” the child’s implementation
- Spawn a child with instructions to do Steps N and N+1
- Mark PLAN checkboxes for steps the child did not complete
- Skip `tests_run` / treat silence as step approval inside the child
- Share one flat `{repo-hash}.json` develop gate across parallel children

After child returns: parent updates `CONTINUITY.md` only (synthesis + paths; keep Memory-bank fields). Then either hand off to a **new chat** for the next step, or ask **sim** again before the next spawn in this conversation - never auto-chain without a gate.

### 6. Safe parallelism (optional)

Default: **serial** - one step in flight (lower context risk).

Parallel Task children **when all** of:

| Condition | Required |
|-----------|----------|
| Distinct PLAN files **or** PLAN explicitly marks steps as independent / parallel-safe | Yes |
| No shared files in step scopes (parent Grep/diff intent) | Yes |
| Deps of each parallel step already complete | Yes |
| User explicitly chose parallel after being asked | Yes |
| **Distinct develop session files** (PLAN-scoped, or PLAN+step when same PLAN) | Yes |

Ask (pt-BR) before any parallel spawn:

```text
Steps {A} e {B} parecem independentes. Executar em paralelo?
(sim / série / cancelar)
```

| Parallel case | Session file per child (`SESSION.md`) |
|---------------|----------------------------------------|
| Different PLANs | `sessions/{repoHash}/plan-{planHash}.json` each |
| Same PLAN, parallel-safe steps | `sessions/{repoHash}/plan-{planHash}-step-{N}.json` each |

Child prompt **must** include: `planPath`, `step`, `memoryBankPath` (read-only), and “load develop SESSION scoped per SESSION.md (PLAN or PLAN+step)”.

If unsure about file independence -> **série**. **Cap: 4** concurrent parallel children; if more steps qualify, wave ≤4 or stay serial. No git worktrees multi-US in MVP (RNF04). Parallelism is supported via scoped sessions - do **not** disable parallel as the only safe path.

**Same-PLAN scope rule:** if any `plan-{planHash}-step-*.json` exists for the PLAN, every spawn for that PLAN (including later serial ones) **must** use PLAN+step session files - do not mix with `plan-{planHash}.json`.

### 7. Stop conditions

Stop spawning and emit handoff when any of:

| Event | Action |
|-------|--------|
| Story PLAN all steps complete | Offer next story or code-review |
| Feature all PLANs complete | Phase -> review; code-review handoff |
| Context pressure (TE02) | Persist CONTINUITY per `context-management.mdc`; resume invoke |
| Context hard-stop | Hard stop; new chat required |
| Child blocked / tests fail | Do not mark step done; report; wait for user |
| User **cancelar** | Leave CONTINUITY with pending next step |

Resume string:

```text
/orchestrate-develop - <full-feature-path>
```

Or per PLAN:

```text
/orchestrate-develop - <full-plan-path>
```

### 8. CONTINUITY

On each meaningful milestone (before/after child, pause, story done):

| Field | Rule |
|-------|------|
| **Phase** | `develop` (or `review` when all done) |
| **Last agent** | `orchestrate-develop` |
| **Memory-bank** | Path + status from Step 0 (`fresh` / `refreshed` / `created`) |
| **Estado atual** | Short per CONTINUITY template: active PLAN, last step done, next step |
| **Handoff tipado** | Exact next `/…` with **full paths** |

Do not paste full diffs, guideline bodies, or memory-bank body into CONTINUITY. CONTINUITY owns phase/handoff; bank does not replace it.

### 9. Step N - Memory Bank refresh-light (after code changes)

When this O3 run had at least one successful develop child that changed application files, **before** the final review handoff:

1. Resolve `bank_root` (same as Step 0).
2. Ask (pt-BR): `Posso atualizar o memory-bank (refresh-light) em '{bank_root}'? (sim / pular / cancelar)`
3. On **sim**: follow `memory-bank-init` mode **`refresh-light`** (inventory + GENERATED + `tech-stack.json` only).
4. Update CONTINUITY Memory-bank status to `refreshed` (or note skipped).
5. On **pular**: log and continue handoff without bank write.

If no app code changed this run, skip Step N.

### 10. Handoff - code-review + manual alternative

When a story or feature develop pass completes (or user asks to review mid-way):

```text
## Handoff O3 -> review

/code-review
/code-review - single
/code-review - multi-angle

## Continuar develop manual (alternativa a O3)
/sdd-develop - <full-plan-path> - Step {N}

## Continuar O3
/orchestrate-develop - <full-feature-path>

## Memory-bank (manual)
/memory-bank-init - refresh-light
```

Suggest `/code-review` (user may pass `- single` or `- multi-angle`; if omitted, **code-review asks**). Never require a mode. O3 does **not** auto-block the pipeline on review.

## Anti-bypass checklist (must enforce)

Parent and children **must not**:

- [ ] Parent implements application/production code or tests
- [ ] One child executes multiple PLAN steps
- [ ] Parent marks steps complete without child success + tests
- [ ] Skip SESSION / user **sim** for the next step spawn
- [ ] Share one flat `{repo-hash}.json` for `step_confirmed` / `tests_run` across parallel children (must use PLAN-scoped or PLAN+step files)
- [ ] Rewrite `sdd-develop` into a looser “fast path”
- [ ] Call `*-developer` from parent to implement the PLAN step (route only via sdd-develop contract in the child)
- [ ] Silence treated as approval

Full copy in `reference.md`.

## Must not

- Skip Step 0 Memory Bank Gate (unless explicit user `skip-memory-bank`)
- Create `memory-bank/` under `features/` or dump bank into CONTINUITY / child prompts
- Parent writes application/production code or tests
- Merge N PLAN steps into one Task / one session context
- Bypass or weaken `sdd-develop` one-step-per-session contract
- Auto-commit / auto-push
- Create external work-item tracker or org-only compliance content
- Force multi-angle code-review
- Introduce git worktrees for multi-US parallelism (MVP)
- Write new PRD/PLAN (O2 / sdd-spec / sdd-plan own that)
- Require memory-bank for manual Classic SDD `sdd-develop` (CA7)
- Pass Task `model` without `SUBAGENT-MODEL.md` gate + user **sim** (or user-named slug); ask model on routine PLAN steps

## Handoff

| Situation | Next |
|-----------|------|
| Next PLAN step | New chat -> `orchestrate-develop` **or** `sdd-develop - <plan> - Step N` |
| Story/feature done | `/code-review` (pass `- single` / `- multi-angle`, or let skill ask) |
| Missing PLAN | `orchestrate-deliver` / `sdd-plan` |
| Prefer no orchestrator | Manual `sdd-develop` only |

### Canonical strings

```text
/orchestrate-develop - <full-feature-path>
```

```text
/sdd-develop - <full-plan-path> - Step N
```

```text
/code-review
/code-review - multi-angle
```

Attribution

tibursocampostibursocampos
View sourceSee grades on GitHubMore from tibursocampos →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Terse caveman voice: answer first, fluff gone, every technical fact kept. Use for /caveman, "caveman mode", "talk like caveman", "be brief", "less tokens". Stays on until "stop caveman" or "normal mode".

1100021 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

698621 votes

Writing Skills

Create and manage Claude Code skills in HASH repository following Anthropic best practices. Use when creating new skills, modifying skill-rules.json, understanding trigger patterns, working with hooks, debugging skill activation, or implementing progressive disclosure. Covers skill structure, YAML frontmatter, trigger types (keywords, intent patterns), UserPromptSubmit hook, and the 500-line rule. Includes validation and debugging with SKILL_DEBUG. Examples include rust-error-stack, cargo-dep...

3931 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3421 votes

catchup

Recovers the conversation and failed tool calls of a previous Codex, Amp, Claude Code, Antigravity, Cline, Copilot CLI, Cursor, DeepSeek Harness, Grok Build, Kimi, OpenCode, Pi Agent, or ZCode session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", asks to recover/summarize a previous session before continuing, or asks to diagnose or report a catchup failure. Do NOT use for the current conversation, git history, or any non-agent log.

741 votes
View all in ai-agents →