| name | autonomous-fleet |
| description | Entry point for the autonomous-fleet multi-agent engineering framework. Use whenever the user wants fully-autonomous coding runs, multi-agent orchestration, PR-per-task pipelines, or mentions autonomous-fleet — even if they have not named a specific mission yet. Routes to one mission or to fleet-program for sequential chains, loads autonomous-fleet-core plus a runtime adapter, and runs unattended on the current repo. Install from github.com/ravidsrk/autonomous-fleet. Trigger on: "use autonomous-fleet", "run autonomous fleet", "autonomous multi-agent run", "fleet mission", "which fleet mission should I use". |
| license | MIT |
| compatibility | Requires git and gh CLI in the target repository; install skills via npx skills |
| metadata | {"author":"ravidsrk","version":"1.0.1","fleet-component":"umbrella"} |
autonomous-fleet
Meta-skill for the autonomous-fleet framework.
This skill does not replace the engine — it orients you, picks a mission, and tells you which
skills to load next.
What this framework is
A skill package of installable skills (this one + core + adapters + 3 proven mission
skills + fleet-program + setup). Each single-mission run composes three layers:
| Layer | Skill(s) | Role |
|---|
| Engine | autonomous-fleet-core | Tool-agnostic coordinator method |
| Adapter | autonomous-fleet-adapter-{orca,claude-code,grok,codex} | Maps primitives to your runtime |
| Mission | doc-sync, test-coverage, adversarial-review-and-fix | Defines the job |
Exploratory missions (not yet shipped end-to-end): see
docs/exploratory/missions/. These are
documented but lack a real-run progress + readiness + external archive
triple and so are not active in the shipped surface.
Install (if skills are not already loaded)
First time on a repo: run /setup-autonomous-fleet after install (adapter, prefix, default bundle).
npx skills add https://github.com/ravidsrk/autonomous-fleet \
--skill setup-autonomous-fleet \
--skill autonomous-fleet-core \
--skill autonomous-fleet-adapter-orca \
--skill fleet-program \
--skill doc-sync \
-y
On Orca, also load Orca's companion skills (orchestration, orca-cli) for full-handoff vs
supervised routing — see autonomous-fleet-adapter-orca → references/orca-platform.md.
For Grok, Claude Code, or Codex hosts, swap -orca for -grok, -claude-code, or -codex.
Install all: npx skills add https://github.com/ravidsrk/autonomous-fleet --skill '*' -y
How to route a request
- Read the user's intent against the mission catalog in references/missions.md.
- If intent names multiple missions, conditional flows ("if audit finds P0…"), or "healthy
repo" → activate
fleet-program (campaign DAG — not several mission skills at once).
- If intent maps clearly to one mission → activate that mission skill and follow it.
- If intent is vague ("clean up this repo") → prefer
fleet-program preset repo-health, or
the closest single mission; Tier 1 for first unattended single-mission runs.
- Always activate
autonomous-fleet-core and one adapter alongside the mission or program.
Default adapter: autonomous-fleet-adapter-orca (reference runtime). On Grok Build without Orca,
use autonomous-fleet-adapter-grok.
Quick routing
| User says | Mission |
|---|
| sync docs, README stale, onboarding wrong | doc-sync |
| add tests, raise coverage | test-coverage |
| audit and fix, harden before prod | adversarial-review-and-fix |
| docs then tests, repo health, mission chain, if-outcome campaign | fleet-program |
| ship safely, harden before PR | fleet-program preset ship-with-proof |
| production ready, quality gate | fleet-program preset quality-gate |
Other intents that previously routed to a dedicated mission
(bug-batch, cleanup, dependency-update, design-integration,
inference-cost, take-product-to-completion, targeted-migration) are
exploratory at the moment — see docs/exploratory/missions/. Parked
mission designs live under docs/exploratory/missions/archive/. Until a
mission is promoted back to skills/, route those intents to
adversarial-review-and-fix (which can file the work as a deferred
mission in its outcome) or fall back to a manual operator pass. Re-route
directly once a mission is promoted.
gstack-derived exploratory missions (2026-06-27) — active designs are
documented in docs/exploratory/missions/, not active until promoted:
| User says | Exploratory mission |
|---|
| browser QA, test the site and fix, dogfood staging | browser-qa-fix |
| incident RCA, root cause, regression test for outage | incident-investigate |
Research: docs/gstack-missions-research.md.
Full tier notes and merge-rate guidance: references/missions.md.
Community skill bundles: autonomous-fleet-core → references/community-skills.md.
Execution checklist
After routing, load and follow these skills in full (do not improvise the method):
autonomous-fleet-core — references/engine.md and references/composition.md- Your runtime adapter skill
- The chosen mission skill or
fleet-program for a chain (one mission active at a time)
The target repo is wherever the user is working (git rev-parse --show-toplevel). No
placeholders — discover maintainer, stack, and scope per the core engine.
Safe defaults
- First unattended run:
doc-sync or test-coverage (Tier 1, highest merge rates).
- Never run
autonomous-fleet-core alone — it needs a mission + adapter.
- Authoring a new runtime: copy
autonomous-fleet-adapter-template, not this skill.
