ccf-pipeline-orchestrator — Claude Code skill: features and install
The ccf-pipeline-orchestrator Claude Code skill plans research stages and coordinates specialist skills. It handles task breakdown, project tracking, and explicitly requested autonomous research.
MIT 2,200 installs ★ 3,769 updated 2026-10-20
npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a claude-codeWhat ccf-pipeline-orchestrator does
The skill maps a project goal to its current stage, assigns specialist owners, and sets gates with evidence and pass conditions. It then integrates their findings. For an explicitly requested autonomous research run, it coordinates the path from a hypothesis and experiment to a supported decision to keep, discard, or branch the approach.
When available, ccfa.yaml supplies project state; the skill can continue from provided artifacts if that file is absent. State changes happen only when project maintenance is authorized. Before planning and handing off work, it activates ccf-humanization and ccf-common; ccf-research-engineer handles implementation and training. Experiments require usable host tools, and a plan or script exit alone does not count as evidence.
The default working location for task files is ccfa-workfiles/<purpose>/<artifact-id>/ under the project root. Planning-only requests produce proposed update instructions rather than writing project state.
What it helps with
- break a research goal into stages and evidence gates
- assign specialist skills to dependent tasks
- track project progress through
ccfa.yaml - combine experiment findings into a next-step decision
Example prompts for the agent
- Break this research goal into stages and assign an owner skill to each task.
- Compare the current artifacts with
ccfa.yamland outline the next steps. - Run an autonomous research cycle within the stated budget and assess the experimental evidence.
How to install ccf-pipeline-orchestrator
1Skills CLI (Vercel)
npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a claude-code2Manually: put the files in place
~/.claude/skills/ccf-pipeline-orchestrator/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/ccf-pipeline-orchestrator/
1Skills CLI (Vercel)
npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a codex2Manually: put the files in place
.agents/skills/ccf-pipeline-orchestrator/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a cursor2Manually: put the files in place
.cursor/skills/ccf-pipeline-orchestrator/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install mikubaka88/CCFA-Skills ccf-pipeline-orchestrator2Manually: put the files in place
.github/skills/ccf-pipeline-orchestrator/SKILL.md1Skills CLI (Vercel)
npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a gemini-cli2Platform CLI
gemini skills install https://github.com/mikubaka88/CCFA-Skills --path ccf-pipeline-orchestratorManually: put the files in place
.devin/skills/ccf-pipeline-orchestrator/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about ccf-pipeline-orchestrator
How do I install ccf-pipeline-orchestrator in Claude Code?
Run npx skills add mikubaka88/CCFA-Skills --skill ccf-pipeline-orchestrator -a claude-code. The skill file is at ~/.claude/skills/ccf-pipeline-orchestrator/SKILL.md.
What does ccf-pipeline-orchestrator do?
It plans CCF research stages, defines evidence gates, and coordinates specialist skills. For an explicit autonomous research request, it also connects hypotheses, experiments, and decisions based on results.
What does an autonomous research run need?
It needs a research objective, a per-run execution contract, and a remaining budget. Experiments also require usable host tools; a plan or successful script exit alone is not experimental evidence.
Source text
SKILL.md · MIT
---
name: ccf-pipeline-orchestrator
description: "Plan or coordinate CCF research stages, goals, gates, artifacts, and ccfa.yaml state. Use for 任务拆解, 流程规划, project status, end-to-end coordination, and explicitly requested autonomous research, autoresearch, or 自主科研迭代. Coordinate hypotheses, engineering, experiments, and feedback within the task budget; research engineer owns code and training."
metadata:
ccf_skill_controls:
handoff_question_mode: partial
respect_session_denylists: true
protect_idea_scope_in_writing: true
private_material_safety: moderate
shared_controls: ../ccf-common/references/
---
# CCF Pipeline Orchestrator
## Family File Contract
Before file work, reuse canonical paths and one task/artifact working directory; default to project-root `ccfa-workfiles/<purpose>/<artifact-id>/`. Update in place, preserve inputs, and use UTF-8 with Chinese-text checks. Apply [artifact-contracts.md](../ccf-common/references/artifact-contracts.md) for naming, ownership, and task-created cleanup; handoffs inherit these paths.
## Collaboration Contract
Before specialist execution, read and apply [ccf-humanization](../ccf-humanization/SKILL.md) first, then [ccf-common](../ccf-common/SKILL.md). At every handoff, reuse their applicable active rules or refresh missing/changed ones. Both preflights are required even without prose; detailed editing, experiment, and maintenance modes run only when relevant.
Integrate necessary specialist contributions under one owner; reuse valid prerequisites and verify affected results. Follow the [cooperation routes](../ccf-common/references/routing.md) without duplicate reports.
## Core Rule
Operate as the project coordinator and workflow planner. Clarify the goal, map the current stage, update or read `ccfa.yaml`, define gates, and name the next owner skill. Specialist skills own downstream outputs. For a plan-only request, return the plan. For explicitly requested end-to-end execution, coordinate the authorized owners through completion rather than stopping after naming the next skill. Follow `../ccf-common/references/task-modes.md`: if the user asks for a short plan, checklist, YAML update, table, or narrative roadmap, use that visible shape instead of forcing a fixed report.
Ensure Humanization and Common are active before planning and every contributing skill. Reuse their applicable rules at handoffs. Apply Humanization's detailed prose modes within authorized writing stages; raw planning and rendering still receive its baseline without a manuscript-edit workflow.
Follow `../ccf-common/references/handoff-modes.md` and `../ccf-common/references/artifact-contracts.md`. Later user corrections normally steer the active project: preserve valid completed work, update affected requirements, and continue. Do not invent completed stages or automatic background jobs.
For an explicit autonomous-research request, load [autonomous-research.md](references/autonomous-research.md) and coordinate the complete authorized loop: scientific objective, candidate mechanism, discriminating experiment, implementation and execution, observed feedback, and a supported keep/discard/branch decision. Use `ccf-research-engineer` for missing codebase, implementation, reproduction, and training expertise, with actual host tools for execution; a plan or script exit alone is not experimental evidence. Reuse existing implementations and measurements when applicable instead of forcing selection, profiling, or retraining before every iteration. The loop may change task-owned research code and search strategies within scope. It does not run for ordinary planning, idea review, or writing, and does not require a new preference switch. Skill-maintenance candidate limits do not cap an explicitly requested research campaign.
## Workflow
1. Identify target venue when relevant, current stage, available artifacts, constraints, deadline pressure, and the user's immediate goal. Distinguish a requested plan, a bounded specialist task, and autonomous research. For the latter, establish its per-run execution contract and remaining budget through `references/autonomous-research.md`; missing execution inputs block only the dependent run, not hypothesis development or preparation.
2. Read `ccfa.yaml` when available; if absent, continue with supplied artifacts. Do not require setup or create project state merely to route work; disclose its absence only when it limits requested tracking.
3. For unclear projects, use `references/workflow-planning/intake-protocol.md`, `approach-options.md`, and `design-brief-template.md`.
4. Assign integrating owners using `../ccf-common/references/routing.md`. Work backward from the requested outcome to missing prerequisites and material checks, including upstream work not named by the user. Group tightly connected decisions under one owner; skill ownership identifies required expertise, not a separate agent for every stage. Delegate complete questions through the shared handoff rules. Distinguish contributions from next-artifact ownership; carry both active preflights and relevant Humanization modes throughout. Preserve explicit scope limits.
5. Define each applicable gate through its required evidence, output, pass condition, blocker, and responsible skill. Mark prerequisites satisfied, missing, conflicting, or outside scope; reuse valid ones and resolve missing/conflicting ones before advancing dependent conclusions. Tool completion alone does not pass a gate.
6. Use `../ccf-common/references/handoff-modes.md` and, for long tasks or recovery, `../ccf-common/references/continuity.md`: carry the current goal, justified route, evidence/version, canonical paths, work status, and exact next action. Reconcile actual artifacts after lost context before resuming or replaying side effects. Apply explicit user changes and evidence-backed corrections within scope. Update existing `ccfa.yaml` stage/gate fields only when state maintenance is authorized; preserve unrelated fields and schema. A planning-only request proposes changes without writing.
7. Integrate specialist results, resolve conflicts against source evidence, and check affected downstream conclusions. Advance gates from actual evidence; preserve valid completed work and continue independent stages around a blocker. Reopen only changed dependencies or unresolved findings. Finish when requested outputs and applicable prerequisites/checks are complete, or identify the precise dependent result that remains incomplete.
## Adaptive Output Contract
Put the requested artifact first: roadmap, next-step decision, task list, handoff packet, or `ccfa.yaml` patch instructions. Use the full structure below only for standard planning, ambiguous multi-stage projects, or when the user asks for a complete coordination report.
For autonomous research, deliver the best supported result or precise unresolved outcome, the mechanisms tried and decision-changing observations, retained artifact/evidence paths, observed resource use, and the stopping reason. Distinguish a completed bounded campaign from a demonstrated scientific claim; a campaign may establish a negative result or leave the goal unresolved. Use the existing task record rather than a report per candidate.
```text
Project goal:
Current stage:
Known artifacts:
Missing artifacts:
Gate decision:
Next owner skill:
Handoff packet:
ccfa.yaml update:
Risks / blockers:
```
## References
- `references/autonomous-research.md`: explicitly requested scientific iteration, execution contracts, strategy updates, evidence-based selection, and recovery.
- `references/workflow-planning/`: project intake, approach comparison, and design briefs when ordinary planning needs them.
© mikubaka88 · MIT · mikubaka88/CCFA-Skills