voodootikigod/adlc/adlc-self-orchestrate
Claude Code and agent skill by voodootikigod/adlc
- community
Checking that the SKILL.md file is still in the repository.
How an Antigravity session decomposes a large build-out into a ticket DAG and drives the agb fleet itself (recursive orchestration). Use when asked to "build X in parallel", "run a fleet", "orchestrate this with agb", or when a task is too large for one session.
Install
npx vskill@latest install voodootikigod/adlc/adlc-self-orchestrateSKILL.md
Self-Orchestration with agb
You can multiply yourself: decompose the work into a ticket DAG, then run agb so parallel agy workers build it while you supervise. Control flow belongs to agb (deterministic); you supply the decomposition and the escalation judgment.
When
Use this only for substantial work (3+ independently-buildable parts). Single bounded tasks: just do them.
Procedure
- Prefer compiling an existing plan. Planning belongs to Antigravity's
plan phase, not to you ad hoc: if this work was planned in a planning conversation (this session or the desktop app), a brain artifact (implementation_plan.md) already exists. Compile it instead of decomposing by hand — npx agb brains to find it, then npx agb plan <id> <repo>, which converts, gates (overlap, coldstart, parallax, premortem), and writes a provenance-stamped plan.json. If gates report blocking findings, refine the plan (the markdown, not the JSON) and recompile. Steps 1–3 below are the fallback for when no brain artifact exists.
- Foundation first. Identify shared surface (schemas, types, shared
utilities). Build and commit it to main BEFORE the fan-out — parallel workers consume the foundation, never invent it. Foundation paths go into every ticket's rails (read-only).
- Decompose into tickets (adlc schema). Each ticket must be
executable by a fresh agent from its body alone — full file paths, exact acceptance criteria, named gate commands. Partition scopes so no two tickets share files; shared needs = a foundation item or an edge.
- Write the plan file
plan.json:
Tier by escape cost, not prestige: cheap (Gemini Flash) where gates catch everything; mid (Claude Sonnet / Gemini Pro) default; frontier only for contracts/migrations with thin rails.
Pool hint routing: pool_hint ∈ gemini | claude | auto. Pair builder pool hints so prosecution runs cross-family (e.g. gemini builder + claude prosecutor) for maximum refutation power.
- Validate, then run:
- Supervise, don't babysit. agb already does two-strike regeneration,
cross-model prosecution, and gate-blocked merges. Your job is only the escalations: tickets that fail twice mean the ticket is wrong — rewrite the ticket (smaller scope, more explicit body), don't coach the agent.
- Report from
.booster/report.json: merged, failed (with reasons),
requests per quota pool.
Quota safety
- Width is capped per pool by agb; do not raise caps above the calibrated
ceilings without re-probing (npx agb probe).
- A run of N tickets costs roughly 2N–4N requests (build + prosecution +
retries). Check that against your weekly budget before launching.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingplugins/adlc-gemini/skills/adlc-self-orchestrate/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/voodootikigod/adlc/adlc-self-orchestrate)