Agent skill

Conductor

by Jinghao67 in Jinghao67/conductor

A skill your agent uses when a complex task needs dependency-aware interactive branches, subagents, or AI coding threads while keeping the master session context clean through explicit branch…

MITAuto-check passedAgent Workflows

Install Conductor

skills CLI
$ npx skills add Jinghao67/conductor --skill conductor -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install Jinghao67/conductor conductor --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/Jinghao67/conductor.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/conductor .claude/skills/conductor && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
conductor
GitHub stars
102
Token cost
~3.9k tokens
SKILL.md length
1,955 words
Files
5 (incl. references)
Skills in repo
2
Repo updated
First seen
Licence
MIT

At a glance

A skill your agent uses when a complex task needs dependency-aware interactive branches, subagents, or AI coding threads while keeping the master session context clean through explicit branch…

  • Works in 5 steps: Keep in master session if it changes… → Use the dispatch session if the… → Create or use an interactive branch if… → …
  • A complex task needs dependency-aware interactive branches
  • SKILL.md covers Purpose, Core Rules, Session Types And Names and Session Registry First, plus 12 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Conductor is an agent skill from Jinghao67/conductor. Use when a complex task needs dependency-aware interactive branches, subagents, or AI coding threads while keeping the master session context clean through explicit branch briefs, completion reports, wave planning, and user-approved merges. Especially useful with grill-me, Trellis, research, implementation, writing, and planning workflows that risk context pollution.

Its SKILL.md is about 3.9k tokens, which your agent loads only when the skill is triggered. The skill folder holds 6 other files, including reference files (for example `agents/openai.yaml`, `references/branch-brief-template.md` and `references/branch-map-template.md`).

It sits in Agent Workflows, covering Requirements gathering. The repository describes itself as: Context conductor for clean master sessions, dirty explainer sidecars, and interactive AI branches. The licence is MIT.

When your agent uses it

  • A complex task needs dependency-aware interactive branches
  • AI coding threads while keeping the master session context clean through explicit branch briefs
  • Completion reports
  • User-approved merges

Example prompts

  • “/conductor”

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

  1. Keep in master session if it changes global goals, scope, constraints, priorities, branch structure, merge decisions, or project snapshots.
  2. Use the dispatch session if the conversation is mainly about whether to open sessions, whether tasks are parallel or serial, dependency…
  3. Create or use an interactive branch if the work is exploratory, implementation-heavy, research-heavy, review-heavy, or likely to produce…
  4. Send to the explainer sidecar if the user wants long-form explanation, background learning, conceptual clarification, or tutorial-style…
  5. Enter merge flow only after a branch is user-confirmed complete.

What it can do on your machine

Read from SKILL.md and the folder at commit b866a50. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md (its code samples are json).

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Conductor loads about 3.9k tokens when it runs, and up to ~5k if it reads all its reference files. Until then it costs about 95 tokens; SKILL.md has 1,955 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~95
When it runs · the whole SKILL.md, loaded when a task matches
~3.9k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~5k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from Jinghao67/conductor at commit b866a50, republished under its MIT licence (© Jinghao67). 1,955 words, ~3,941 tokens.

Download SKILL.mdSave it as .claude/skills/conductor/SKILL.md (or your agent's skills folder). This skill also uses 4 other files; get the full folder from GitHub.
name
conductor
description
Use when a complex task needs dependency-aware interactive branches, subagents, or AI coding threads while keeping the master session context clean through explicit branch briefs, completion reports, wave planning, and user-approved merges. Especially useful with grill-me, Trellis, research, implementation, writing, and planning workflows that risk context pollution.

Conductor

Purpose

conductor is a context isolation and branch registry protocol. It keeps one master session clean while messy exploration, implementation, research, review, or learning happens in separate user-interactive branch sessions.

Use it to decide whether a request belongs in the master session, an interactive branch, an explainer sidecar, or a merge flow. It does not replace Trellis executors or other workflow agents; it governs how branches are created, tracked, completed, and optionally merged.

Core Rules

  • The master session owns the project overview: goal, constraints, branch map, global decisions, approved summaries, risks, and next steps.
  • Use session registry first. Create visible branch cards before creating real sessions.
  • Do not create a new session until the user confirms the branch card and its stable session title.
  • Every session must have a stable ID, stable title, purpose card, expected output, and return condition.
  • Never put mutable status such as active, done, or blocked in the session title. Status belongs in the branch map or Today View.
  • Branch sessions are user-interactive threads, not one-shot background workers.
  • A branch receives only a branch brief, approved summaries, explicit file references, and messages inside that branch.
  • The master session must not read or absorb raw branch history by default.
  • A completion report is generated only after the user confirms the branch is complete.
  • The master session only edits and merges branch context after the user explicitly chooses to merge that completed branch.
  • Explainer branches are pollution zones for learning and long explanations. They default to no merge.
  • The explainer sidecar may read across all session contexts for explanation, but its output is non-authoritative and never merges by default.
  • Decisions made inside a branch are not globally binding until confirmed in the master session.
  • Do not assume all branches can run in parallel. Always run a dependency pass before opening branches.

Session Types And Names

Use exactly these session types:

TypeStable title patternPurposeDefault merge behavior
master[CD-MAIN][master] <project>Project control room: global goals, branch map, confirmed decisions, risks, approved summariesSource of truth
dispatch[CD-DISPATCH][routing] Branch planningDiscuss whether to open sessions, whether work is parallel or serial, dependency waves, and session cardsMerge only final dispatch decisions
branch[CD-001][W1][task] First confirmed branchUser-interactive work session for one bounded taskReport generated after user-confirmed completion; merge only after explicit user approval
explainer[CD-E01][sidecar][explainer] Dirty questionsDirty learning session for questions, tutorials, and cross-session explanationNo merge by default

Rules:

  • Session titles must start with the Conductor ID.
  • Session titles must include role and wave when relevant.
  • Session titles must be short enough to identify in a thread list.
  • Do not create multiple explainer sessions. Use one fixed explainer per project.
  • Do not create multiple dispatch sessions. Use one fixed dispatch session per project.
  • Prefer at most 2 active interactive branch sessions, plus the optional dispatch session and fixed explainer sidecar.

Session Registry First

Conductor creates branch cards before sessions. A branch card must include:

  • ID and stable session title
  • type and role
  • one-sentence purpose
  • why it exists
  • dependency/wave placement
  • allowed context
  • expected artifact
  • completion criteria
  • return condition
  • whether it should open now, stay planned, or stay blocked

Only after the user confirms the card should Conductor create or bind a real session. When entering a branch, the first visible content must be a Purpose Card:

text
You are branch CD-001: First confirmed branch
Purpose: do the first bounded task the user confirmed
Not for: unscoped work outside the branch card
Input: master snapshot snap-xxx; approved summaries only
Output: completion-report.md
Return to master when: branch output is ready and the user confirms completion

After creating, blocking, completing, merging, parking, or archiving a session, refresh a compact Today View in the master session:

text
Active now:
- CD-001 First confirmed branch — task — waiting for user review
- CD-E01 Dirty questions — sidecar explainer

Planned, not opened:
- CD-002 Implementation plan — waits for CD-001

Merge pending:
- none

Routing

Whenever this skill is active, route each new request before acting:

  1. Keep in master session if it changes global goals, scope, constraints, priorities, branch structure, merge decisions, or project snapshots.
  2. Use the dispatch session if the conversation is mainly about whether to open sessions, whether tasks are parallel or serial, dependency order, wave planning, or branch-card design.
  3. Create or use an interactive branch if the work is exploratory, implementation-heavy, research-heavy, review-heavy, or likely to produce many intermediate details.
  4. Send to the explainer sidecar if the user wants long-form explanation, background learning, conceptual clarification, or tutorial-style help.
  5. Enter merge flow only after a branch is user-confirmed complete.

Keep this routing lightweight. Do not turn every message into process ceremony.

Dispatch Session

Do not open the dispatch session by default. Suggest opening [CD-DISPATCH][routing] Branch planning when:

  • branch planning takes more than 2-3 turns
  • there are more than 3 candidate branch cards
  • dependency order is unclear
  • the user asks whether work should be parallel or serial
  • session proliferation is starting to make the project hard to navigate

The dispatch session is isolated from the master session. It may discuss:

  • which sessions should exist
  • which branches should be planned, opened, parked, blocked, or archived
  • which work can run in parallel and which must be serial
  • branch titles, roles, expected outputs, and completion criteria
  • active-session budget

The dispatch session must not do implementation, research, review, or long-form explanation. It must not read raw branch histories by default. It returns only a compact dispatch decision to the master session, for example:

text
Dispatch decision:
- Open CD-001 and CD-E01 now.
- Keep CD-002 planned until CD-001 completion report is ready.
- Do not open implementation branch yet.

Dependency Pass And Wave Plan

Before creating or opening branches, identify execution order. Many branches depend on outputs from earlier branches and must remain planned or blocked until their prerequisites are done.

Classify each proposed branch:

  • ready_parallel: can start now from the current master snapshot without waiting on another branch.
  • dependent: needs a specific branch output, decision, artifact, or user confirmation first.
  • gate: a review, merge, or master-session decision required before the next wave can start.
  • optional: useful but not on the critical path.
  • explainer: can run as a sidecar unless the user explicitly makes it blocking.

Output a compact wave plan before creating threads:

WaveBranchesPrerequisitesGate to unlock next wave
0master decisionsnoneuser confirms scope
1branches that can run nowcurrent snapshotcompletion reports reviewed
2dependent branchesWave 1 outputsmerge or user decision

Rules:

  • Only branches in the current unlocked wave should become active by default.
  • Later-wave branches may be recorded as planned; mark them blocked if the user tries to start them before prerequisites are met.
  • Parallel means "safe to run from the same snapshot without needing each other's outputs", not merely "different topics".
  • If dependencies are ambiguous, ask the user to confirm the order before creating branch threads.
  • After a wave completes, update the snapshot, note what it unlocks, and then propose the next wave.

Branch State Machine

Use these stable states:

StateMeaningTypical next step
plannedSuggested by master, not confirmedcreate brief or cancel
brief_readyBranch brief exists, thread not createdcreate thread or revise brief
activeUser can enter and interactblock, park, or suggest completion
blockedWaiting on master, user, external input, or another branchresume or cancel
completion_suggestedBranch thinks it may be doneuser confirms completion or continues
report_readyUser confirmed completion and report existsrequest merge or archive
merge_pendingMaster asks whether to merge reportmerge, reject, or defer
mergedMaster wrote approved compressed recordarchive
rejectedUser chose not to mergearchive
archivedHidden from active viewreopen

Default active interactive branch limit: 2. The fixed explainer sidecar and optional dispatch session do not count as interactive branches. If creating a third active branch, suggest parking, completing, or archiving an existing branch first.

Show full SKILL.md (752 more words)Show less

Creating Branches

Create branches only after user confirmation. First show:

  • stable session title
  • proposed branch title and role
  • why it should be a branch
  • whether it is in the current wave or waiting on prerequisites
  • dependencies, unblocks, and gate condition
  • expected artifact
  • completion criteria
  • whether it needs a Trellis child task, an AI coding thread, or both

Create only current-wave branches by default. Do not create all planned branch threads at once unless the user explicitly requests that.

After confirmation, write a branch brief using references/branch-brief-template.md. If thread tools are available, create or bind an AI coding thread, set the thread title to the stable session title when possible, and give it only the branch brief as the initial prompt. Record the returned thread_id; a Codex deep link such as codex://threads/<thread-id> is optional when using Codex and is not the source of truth.

If thread creation is unavailable, create the brief and ask the user to start a separate session with it.

Completion And Merge

A branch may suggest completion, but only the user can confirm it. After confirmation, generate a completion report using references/completion-report-template.md.

The master session then asks whether to merge:

  • Merge: read the completion report, compress it, normalize terminology, and add a short approved record to the master context.
  • Reject: record that the branch completed but was not merged.
  • Defer: leave the branch in merge_pending.

Do not read raw branch history unless the user explicitly asks to debug or audit that branch. The default merge input is completion-report.md, especially its Suggested Merge Note.

Explainer Sidecar

Use one explainer thread per project. It is for questions that would heavily pollute context: tutorials, long derivations, conceptual gaps, background knowledge, and repeated clarification.

The explainer is context-rich but non-authoritative. It may read across all session contexts to answer user questions, but should load only the relevant context for the current question and should label source authority:

  • L0: master snapshot, branch map, Today View - confirmed
  • L1: branch cards and branch briefs - intended scope
  • L2: completion reports and approved summaries - reviewed outputs
  • L3: raw branch histories - branch-local and unconfirmed; read only when the user explicitly asks
  • L4: explainer history - dirty learning context

Do not create a Trellis child task for the explainer by default. Represent it as a sidecar in the branch map. Only promote explainer output if the user strongly requests it and the content becomes a project decision, constraint, terminology definition, or spec/task update. Even then, merge as a very short project knowledge item, not as teaching material.

Global Decisions

If a branch discussion creates or implies a decision affecting other branches, architecture, scope, naming, deadlines, acceptance criteria, or project goals, the branch must record it as a proposed global decision. It is not binding until the master session confirms it.

Completion reports should include proposed global decisions with:

  • decision
  • affected scope
  • why it matters
  • affected branches
  • recommended master-session action

Snapshots And Staleness

The master session should create short snapshots after major events: branch creation, branch completion, merge, archival, and confirmed global decisions.

Each branch brief should record:

  • based_on_snapshot_id
  • brief_version
  • creation date
  • parent task or root branch

When the master snapshot changes in a way that affects an active branch, mark that branch stale and ask whether to refresh its brief. Do not silently sync hidden context.

Trellis Best Practice

With Trellis, map conductor like this:

  • master session: parent/root task
  • dispatch session: sidecar AI coding thread for branch planning; not a Trellis child task by default
  • interactive branch: Trellis child task plus a user-enterable AI coding thread
  • explainer: sidecar AI coding thread, not a Trellis child task by default
  • branch map: parent task branch-map.md
  • machine binding: task.json.meta.conductor

Prefer Trellis scripts such as task.py create --parent or task.py add-subtask for parent/child relationships. Do not manually edit parent or children unless Trellis scripts are unavailable or fail.

Use task.json.meta.conductor only for small machine-readable fields:

json
{
  "conductor": {
    "branch_id": "CD-001",
    "branch_type": "branch",
    "interaction_mode": "user_interactive",
    "role": "research",
    "parent_branch_id": "CD-MAIN",
    "thread_id": "thr_xxx",
    "branch_brief_path": ".trellis/tasks/example/branch-brief.md",
    "completion_report_path": ".trellis/tasks/example/completion-report.md",
    "status": "active",
    "brief_version": 1,
    "based_on_snapshot_id": "snap-2026-06-03-001",
    "execution_wave": 1,
    "depends_on": [],
    "unblocks": ["CD-002"],
    "start_policy": "current_wave_only",
    "gate_condition": "completion_report_ready",
    "merge_policy": "explicit_user_confirm"
  }
}

Do not add branch briefs, completion reports, or raw conversation summaries to implement.jsonl or check.jsonl by default. Those files are for Trellis execution/check context and should be touched only when the user confirms the branch artifact must become implementation or check context.

Non-Trellis Fallback

If Trellis is not available, maintain:

  • conductor.yaml as the machine-readable registry
  • branch-map.md as the human-readable snapshot
  • Mermaid inside branch-map.md for visualization

Use references/branch-map-template.md when creating or refreshing the map.

When To Stop And Summarize

When the user asks for a Trellis-ready summary or before implementation begins, provide:

  • goal and non-goals
  • user flow
  • routing rules
  • branch states
  • Trellis mapping or fallback registry
  • acceptance criteria
  • risks
  • open questions

© Jinghao67, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 4 other files (references) in skills/conductor of Jinghao67/conductor.

  • SKILL.md
  • agents/openai.yaml
  • references/branch-brief-template.md
  • references/branch-map-template.md
  • references/completion-report-template.md

Open the folder on GitHubat commit b866a50

Compare with similar skills

Conductor next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Conductor compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Conductor this skillJinghao67/conductor102—~3.9kAutomated safety check: PassMIT
Using Superpowersfarm-fe/farm5.6k35 repos~1.4kAutomated safety check: PassMIT
Interview Meaddyosmani/agent-skills103k6 repos~3.8kAutomated safety check: PassMIT
Grillingpietheinstrengholt/rssmonster56432 repos~510Automated safety check: PassMIT
Agentic Workflow Designerdotnet/Open-XML-SDK4.6k2 repos~3.5kAutomated safety check: PassMIT
Ask User QuestionMemTensor/MemOS12k—~1kAutomated safety check: PassApache-2.0

Similar skills

  • Using Superpowers

    farm-fe/farm

    A skill your agent uses when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions

    5.6k GitHub starsUsed in 35 repos~1.4k tokens
    Agent WorkflowsAuto-check passed
  • Interview Me

    addyosmani/agent-skills

    Asks one question at a time, each with a best guess attached, until the agent is about 95 percent sure what you really want, before any plan, spec or code.

    103k GitHub starsUsed in 6 repos~3.8k tokens
    Agent WorkflowsAuto-check passed
  • Grilling

    pietheinstrengholt/rssmonster

    Grill the user relentlessly about a plan, decision, or idea.

    564 GitHub starsUsed in 32 repos~510 tokens
    Agent WorkflowsAuto-check passed
  • Agentic Workflow Designer

    dotnet/Open-XML-SDK

    Official

    Interviews you one question at a time about goal, trigger, permissions and data needs, then drafts a single agentic workflow markdown file.

    4.6k GitHub starsUsed in 2 repos~3.5k tokens
    Agent WorkflowsAuto-check passed
  • Ask User Question

    MemTensor/MemOS

    Shows a question as a modal in the interface to clarify a task, collect a preference or get approval, since the user cannot see terminal output.

    12k GitHub stars~1k tokensUpdated 9 days ago
    Agent WorkflowsAuto-check passed
  • Brainstorming Before Building

    jnMetaCode/superpowers-zh

    Turns a rough idea into an approved design before any code is written, sorting the request into spike, bounded or architectural and enforcing an approval gate.

    8.3k GitHub stars~1.8k tokensUpdated 4 days ago
    Agent WorkflowsAuto-check passed

More from Jinghao67/conductor

  • Conductor Cc

    Jinghao67/conductor

    A skill your agent uses when a complex, long-running task risks polluting one conversation — research, implementation, review, planning, or learning that should be split into dependency-aware…

    102 GitHub stars~1.1k tokensUpdated 4 mo ago
    Auto-check passed

Categories

Questions about Conductor

What does Conductor do?

A skill your agent uses when a complex task needs dependency-aware interactive branches, subagents, or AI coding threads while keeping the master session context clean through explicit branch…. Conductor is an agent skill from Jinghao67/conductor. Use when a complex task needs dependency-aware interactive branches, subagents, or AI coding threads while keeping the master session context clean through explicit branch briefs, completion reports, wave planning, and user-approved merges.

When should I use Conductor?

Conductor fits situations like: A complex task needs dependency-aware interactive branches; AI coding threads while keeping the master session context clean through explicit branch briefs; completion reports; user-approved merges.

How do I install Conductor in Claude Code?

Run `npx skills add Jinghao67/conductor --skill conductor -a claude-code`. Or copy the skill folder (skills/conductor in Jinghao67/conductor) into .claude/skills/conductor in your project. Claude Code loads it when a task matches its description.

How do I install Conductor in Codex?

Run `npx skills add Jinghao67/conductor --skill conductor -a codex`. Or copy the skill folder (skills/conductor in Jinghao67/conductor) into .agents/skills/conductor in your project. Codex loads it when a task matches its description.

Can I use Conductor in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add Jinghao67/conductor --skill conductor -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/conductor, .gemini/skills/conductor, .github/skills/conductor and .opencode/skills/conductor in your project.

What does Conductor need to run?

SKILL.md names no scripts, command-line tools or credentials: Conductor is instructions for the agent only.

Does Conductor access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Conductor safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Conductor use?

Conductor is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Conductor use?

About 3.9k tokens (SKILL.md is roughly 16k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 1.1k tokens, read only when the agent opens those files.

What are the alternatives to Conductor?

Skills that share tags, products or a category with Conductor: Using Superpowers (farm-fe/farm, 5.6k stars), Interview Me (addyosmani/agent-skills, 103k stars), Grilling (pietheinstrengholt/rssmonster, 564 stars) and Agentic Workflow Designer (dotnet/Open-XML-SDK, 4.6k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Conductor?

Jinghao67 (a GitHub user) maintains it in Jinghao67/conductor, which has 102 GitHub stars. The repository holds 2 skills in this directory. The repository was last updated on June 10, 2026.

Source: Jinghao67/conductor on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.