orchestrate
Conduct an existing GitHub spec through an accepted ticket graph, implementation, integration, and cleanup.
Skill by giacomoguidotto
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/giacomoguidotto-orchestrate
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
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
- Runs as
- Node.js package
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/orchestrate
What's inside
- Skill name
orchestrate- Description
- Conduct an existing GitHub spec through an accepted ticket graph, implementation, integration, and cleanup.
- License
- MIT
- Instructions
- 190 lines in SKILL.md · view on GitHub
Before installing
- Ships executable scripts
Files (16)
- SKILL.md
- openai.yaml · agents/
- IMPLEMENTER.md · references/
- RECOVERY.md · references/
- REVIEW.md · references/
- REVIEWER.md · references/
- RUNTIME.md · references/
- conductor-state.mjs · scripts/
- conductor-state.test.mjs · scripts/
- liveness-contract.test.mjs · scripts/
- render-graph.mjs · scripts/
- render-graph.test.mjs · scripts/
- upsert-graph.mjs · scripts/
- upsert-graph.test.mjs · scripts/
- validate-graph.mjs · scripts/
- validate-graph.test.mjs · scripts/
SKILL.md (first lines)
# Orchestrate
Act as a **conductor**. The spec and implementation tickets already exist. This
task proves their graph, opens one launch gate, publishes the accepted graph,
then starts exactly one lifecycle actor.
Accept `spec <issue-ref>` plus an optional
`owner/repo:final-branch:integration-branch`. `#1234` uses the current GitHub
repository; a full issue URL resolves its repository. A named profile is a
recommendation and never skips the launch gate.
## Profile
The launch profile has two orthogonal axes:
- `review=lean`: required validation plus local CodeRabbit light review.
- `review=deep`: required validation plus full local CodeRabbit and one
independent adversarial reviewer.
- `supervision=supervised`: integration pauses for SHA-bound user approval.
- `supervision=unsupervised`: a ready SHA integrates automatically.
Repository-required checks are always additive. Delivery is automatic: an
exclusive writable lane may integrate without a PR; parallel work or repository
policy uses PRs. Worker execution permission is not a supervision axis: every
profile requires writable actor worktrees and non-interactive execution.
Keep preflight read-only. The accepted launch gate authorizes publication of the
managed graph section, lifecycle actors, assignments, branches, PRs, comments,
admitted integration, issue closure, and exact cleanup. Unsupervised acceptance
also authorizes corrective implementation tickets, blocker edges, and PATCH
releases required to satisfy the accepted spec. It does not authorize changed
acceptance criteria, a MINOR or MAJOR release, new external side effects, or an
irreversible destructive choice.
## 1. Establish the score
1. Treat the issue repository as the ledger repository. Read its instructions
and resolve its local checkout and Codex project.
2. Fetch the full spec body and comments. Resolve the assignee from an explicit
invocation override or the authenticated GitHub viewer.
3. Build the complete implementation-ticket inventory from derivation links,
comments, every issue whose `Parent` names the spec, and native sub-issues.
Fetch native sub-issues independently; neither source alone is the inventory.
Fetch every expected ticket's body, comments, state, and linked PRs. Precursors
and decision issues do not enter the implementation manifest.
4. Require every expected implementation ticket to be a native sub-issue. List
every missing relationship and stop before rendering when any is absent.
5. Classify a ticket as `HITL` when its labels, metadata, or body explicitly say
`HITL`, `HILT`, or human-in-the-loop; otherwise classify it as `AFK`.
6. Resolve `Blocked by` edges from native relationships, then ticket text.
Record repository-qualified external blockers separately.
7. For each ticket, distinguish its ledger repository from its implementation
target. Resolve an explicitly named target repository from ticket metadata,
body, or accepted spec evidence; otherwise use the ledger repository. Record
its target branch, whether it has any refs, local checkout and Codex project
when available, and target-specific validation command. Never infer that a
native sub-issue implements in its ledger repository.
8. Resolve final and integration branches per implementation target from ticket
evidence, invocation, an unambiguous target convention, then its default
branch. Resolve whether target rules or PR-only checks require PR delivery.
9. Prove the current trusted runtime can create writable isolated worktrees and
perform required git and GitHub operations without operational approval.
A runtime that would prompt during implementation is not launchable.
10. Find existing conductors and live work for this spec. Record their profile,
target branch, ticket, PR, head, and state without resuming them. A legacy
conductor without a profile is `deep + supervised`.
Completion criterion: at least one implementation ticket exists; each has a
title, state, mode, native parent, resolved blockers, and implementation target;
the ledger repository, assignee, target branches, validation commands, PR
requirements, and every live actor are known.
## 2. Prove the frontier
Build the manifest from live GitHub state:
```json
{"spec":"1234","ledgerRepository":"owner/ledger","tickets":[{"id":"10","parent":"1234","nativeSubIssue":true,"title":"...","state":"open","mode":"hitl","blockedBy":["other/repo#7"],"targetRepository":"owner/code"}],"externalBlockers":[{"id":"other/repo#7","state":"closed"}]}
```
Resolve `scripts/validate-graph.mjs` beside this file and run it with `node`,
passing the manifest on stdin. A nonzero exit stops preflight. Repair discovery
from authoritative state; if it remains invalid, report the diagnostic and end.
Run `scripts/render-graph.mjs` with the same manifest. Keep its Mermaid stdout,
`launchable` AFK frontier, and `hitlFrontier` pauses. Mark the current frontier
`direct-eligible` only when it contains one AFK ticket, no other live writer
targets integration, direct pushes are allowed, and complete validation runs
before integration. Uncertain or PR-only validation makes it `pr-required`.
Completion criterion: validation proves a complete acyclic graph and the
renderer emits one node per ticket and external blocker; current delivery
eligibility has evidence.
## 3. Open the launch gate
Show the Mermaid graph and say arrows point from blocker to unlocked ticket.
List every open HITL ticket, or `HITL pauses: none`. Show current delivery
eligibility and any existing conductor beside the graph.
Ask exactly: "Which orchestration profile should I launch? Choosing one accepts
this graph, its listed HITL pauses, and the repair authority described above."
1. `lean + supervised`
2. `lean + unsupervised`
3. `deep + supervised`
4. `deep + unsupervised`
If a live conductor differs, state that selection authorizes replacement only
after exact cleanup proves no unmerged work will be lost. Create nothing and
write nothing before the answer.
Completion criterion: the user selected one profile and accepted the graph and
HITL pauses in the same answer.
## 4. Publish the accepted graph
Refresh the spec, tickets, blockers, implementation targets, branches, and
repository rules. Rebuild, validate, and render the manifest. Structural orFrom giacomoguidotto/skills, MIT.
More from giacomoguidotto/skills
- 0.546score
setup-agentic-os
giacomoguidotto
Check or reconcile the fixed Agentic Constellation through the three System-owned setup interfaces.
SkillBFreshness: B★ 1Claude Code +9 - 0.502score
setup-knowledge-system
giacomoguidotto
Check or reconcile the standalone Knowledge System.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
domain-reconnaissance
giacomoguidotto
Domain reconnaissance exposes authoritative constraints around a high-cost design decision.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
lookup
giacomoguidotto
Look up scoped live Knowledge Bank context without writing.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
post
giacomoguidotto
Draft or revise ready-to-copy social posts without scheduling or publishing them.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
tweet
giacomoguidotto
Draft or revise ready-to-copy tweets without scheduling or publishing them.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
capture
giacomoguidotto
Capture durable conversation or session knowledge into the Knowledge Bank.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
agentic-os
giacomoguidotto
Coordinate versioned cross-System capabilities.
SkillCFreshness: C★ 1Claude Code +9 - 0.496score
setup-career-system
giacomoguidotto
Check or reconcile a standalone Career System checkout through its canonical versioned gateway.
SkillCFreshness: C★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 0
- Open issues
- 1
- Last commit (this folder)
- Jul 24, 2026 (3 months ago)
- Last release
- Jul 24, 2026 (3 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/giacomoguidotto/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from giacomoguidotto
- 0.546score
setup-agentic-os
giacomoguidotto
Check or reconcile the fixed Agentic Constellation through the three System-owned setup interfaces.
SkillBFreshness: B★ 1Claude Code +9 - 0.502score
setup-knowledge-system
giacomoguidotto
Check or reconcile the standalone Knowledge System.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
capture
giacomoguidotto
Capture durable conversation or session knowledge into the Knowledge Bank.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
domain-reconnaissance
giacomoguidotto
Domain reconnaissance exposes authoritative constraints around a high-cost design decision.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
lookup
giacomoguidotto
Look up scoped live Knowledge Bank context without writing.
SkillCFreshness: C★ 1Claude Code +9 - 0.497score
post
giacomoguidotto
Draft or revise ready-to-copy social posts without scheduling or publishing them.
SkillCFreshness: C★ 1Claude Code +9
Cite this page
orchestrate skill. Agent Mart. https://agentmart.dev/skills/giacomoguidotto-orchestrate (last updated 2026-10-24)