autonomous-orchestration
Use when the user explicitly hands over multi-PR coordination — asking Claude to follow through on a user story, work bundle, or backlog scope autonomously…
Skill by els0r
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/els0r-autonomous-orchestration
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, devops, productivity, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/autonomous-orchestration
What's inside
- Skill name
autonomous-orchestration- Description
- Use when the user explicitly hands over multi-PR coordination — asking Claude to follow through on a user story, work bundle, or backlog scope autonomously over multiple turns. Spans issue triage, subagent dispatch in isolated worktrees, PR/CI watching, and queueing follow-on issues. Example triggers — "work this autonomously", "follow through on this bundle", "orchestrate this story", "self-drive #N", "run the backlog from X to Y", "take this end-to-end". SKIP for — single-PR work, XS changes, direct one-shot edits, code review of an existing diff, bug-fix triage, doc tweaks, or any task where the user is steering each step. If unsure whether the scope qualifies, do NOT invoke — ask the user whether they want autonomous mode before proceeding.
- License
- MIT
- Instructions
- 99 lines in SKILL.md · view on GitHub
Files (4)
- SKILL.md
- dispatch.md
- handover-points.md
- recovery.md
SKILL.md (first lines)
# Autonomous orchestration
Operate as an orchestrator across several issues, PRs, and a
long-running session — not as a single-PR contributor.
**Preconditions.**
- The user has explicitly handed over coordination (see SKIP list
above; when in doubt, ask before invoking).
- If the project has a working agreement / contribution guide
(sizing, labels, the `Closes #N` convention, worktree practices),
follow it. This skill adds orchestration behavior on top.
## Mode of operation
Run on a dedicated session branch (e.g.
`claude/autonomous-<purpose>-<id>`). It is a coordination workspace
only: it holds no code commits and never opens its own PR — if the
orchestration itself needs an artifact (script, doc), file an XS PR
for it on its own branch off `main`. Implementation happens on
per-issue feature branches cut from `main`, in isolated worktrees,
dispatched to subagents.
**The loop:**
1. Triage the user story into issues (`gh issue create`), applying
the project's triage taxonomy (size, priority, component, kind)
if it has one.
2. Confirm bundle composition with the user before dispatching any
multi-issue agent ([handover-points.md](handover-points.md)).
3. For each issue, dispatch an implementation agent — read
[dispatch.md](dispatch.md) first to select the backend (container
layer vs `Agent` worktree) and build a self-contained brief.
4. Subscribe to the resulting PR via the GitHub MCP server (or
equivalent) so CI and review events stream back.
5. When the agent reports done, verify against GitHub facts, not its
summary ([dispatch.md](dispatch.md) § After the run).
6. On merge, queue the next dependent issue. On plan invalidation or
scope surprises, see [recovery.md](recovery.md).
## Guardrails — never autonomous
Always in force, for the orchestrator and every subagent. Each
requires explicit per-action user authorization; an authorization for
one instance does not generalize. See [dispatch.md](dispatch.md) for
how subagents inherit them.
- **No `git push --force`** without permission; to `main` / release
branches, even with permission, push back and propose a
non-destructive alternative.
- **No `git reset --hard`, `git clean -fd`, `git checkout --`** on
branches with unpushed work or unfamiliar files. Investigate
unexpected state before deleting it — it may be the user's WIP.
- **No amending pushed commits.** Fix forward with a new commit,
never `git commit --amend` on a published SHA.
- **No `--no-verify`, `--no-gpg-sign`, or other hook-skip flags.** A
failing hook is a signal, not an obstacle.
- **No merge, auto-merge, review-request, or PR-close.** The
orchestrator opens and watches PRs; merging is always the user's
action.
- **No direct issue close.** Let `Closes #N` in a merged PR do it.
- **No branch / worktree deletion** without asking, however stale
the branch looks.
- **Implementation branches off `main`**, never off the
orchestration branch.
If a guardrail blocks progress, stop and ask — it is a handover
point. Never search for a flag that bypasses it.
## When to ask the user
Use `AskUserQuestion` only at the six handover points in
[handover-points.md](handover-points.md): bundle composition, plan
invalidation, trade-offs with no clean default, destructive recovery,
post-merge re-dispatch, and guardrail blocks. Routine progress, CI
greens, and webhook events do **not** require asking. Read that file
before asking — and whenever a system event (webhook, task
notification, session resume) tempts you to act on it as if it were
user input.
## Quick checklist (before dispatching an autonomous bundle)
- [ ] User story decomposed into issues, project triage taxonomy
applied if any.
- [ ] Bundle composition confirmed with the user.
- [ ] Dispatch backend selected per [dispatch.md](dispatch.md) — the
project's container layer when present, host `Agent` worktree
otherwise.
- [ ] Each subagent brief is self-contained (issue verbatim, prior
decisions, branch/base, commit convention, PR requirements,
guardrails).
- [ ] Implementation branches cut from `main`, not the orchestration
branch; isolated worktrees.
- [ ] PRs subscribed via the GitHub MCP server on open.
- [ ] Subagent reports verified via `gh` / GitHub MCP before relaying
to the user.
- [ ] No guardrail action taken without per-action user OK.From els0r/skills, MIT.
More from els0r/skills
- 0.547score
blast-radius
els0r
Produce a risk-ranked audit map of a codebase (especially AI-generated code the user hasn't read), so the human can spend limited reading time where failure…
SkillBFreshness: B★ 1Claude Code +9 - 0.497score
go-testing
els0r
Go testing standards: table-driven tests, testify/require, sentinel errors, mocking, fuzzing, race detection, and test file conventions.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
design-rubric
els0r
Use when critiquing, reviewing, or refining visual designs, wireframes, or UI layouts — especially for small/medium SPAs.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
go-instrumentation
els0r
Observability principles for Go services: structured logging, distributed tracing, metrics, context propagation, and HTTP service bootstrap.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
go-performance
els0r
Go performance analysis: memory allocation, CPU bottlenecks, concurrency contention, struct alignment, escape analysis, and benchmarking patterns.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
go-style
els0r
Go coding conventions, error handling, interface design, and idiomatic patterns.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
improve-go-architecture
els0r
Find deepening opportunities in a Go codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/.
SkillCFreshness: C★ 1Claude Code +9 - 0.494score
coding-style
els0r
Use for all code generation, review, and refactoring regardless of language.
SkillCFreshness: C★ 1Claude Code +9 - 0.486score
Codebase Locator
els0r
Locates files, directories, and components relevant to a feature or task.
SkillCFreshness: C★ 1Claude Code +9 - 0.486score
Codebase Analyzer
els0r
Analyzes codebase implementation details.
SkillCFreshness: C★ 1Claude Code +9 - 0.472score
grill-with-docs
els0r
Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as…
SkillCFreshness: C★ 1Claude Code +9 - 0.468score
diagnose
els0r
Disciplined diagnosis loop for hard bugs and performance regressions.
SkillCFreshness: C★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Jun 9, 2026 (4 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/els0r/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from els0r
- 0.547score
blast-radius
els0r
Produce a risk-ranked audit map of a codebase (especially AI-generated code the user hasn't read), so the human can spend limited reading time where failure…
SkillBFreshness: B★ 1Claude Code +9 - 0.497score
go-performance
els0r
Go performance analysis: memory allocation, CPU bottlenecks, concurrency contention, struct alignment, escape analysis, and benchmarking patterns.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
go-style
els0r
Go coding conventions, error handling, interface design, and idiomatic patterns.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
design-rubric
els0r
Use when critiquing, reviewing, or refining visual designs, wireframes, or UI layouts — especially for small/medium SPAs.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
go-instrumentation
els0r
Observability principles for Go services: structured logging, distributed tracing, metrics, context propagation, and HTTP service bootstrap.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
improve-go-architecture
els0r
Find deepening opportunities in a Go codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/.
SkillCFreshness: C★ 1Claude Code +9
Cite this page
autonomous-orchestration skill. Agent Mart. https://agentmart.dev/skills/els0r-autonomous-orchestration (last updated 2026-10-24)