claude-output-lane
Routes reusable Claude output (plans, ideas, findings, reviews, draft artifacts, diff proposals, questions, decisions, evidence packets) for Codex/main inspection. Use when Claude produces structured output that another lane must inspect, or when mentioning 'Claude output', 'lane handoff', 'output routing', or 'cross-lane artifact'.
- License 许可
- MIT
- 最近更新
- 2026-11-11
- Stars
- ★ 2
- Forks
- 0
01 前言 Front Matter
- name
- claude-output-lane
- scope
- harness
- triggers
- [ "claude output lane", "reusable review output", "draft artifact lane" ]
- invocation
- ability
- description
- Routes reusable Claude output (plans, ideas, findings, reviews, draft artifacts, diff proposals, questions, decisions, evidence packets) for Codex/main inspection. Use when Claude produces structured output that another lane must inspect, or when mentioning 'Claude output', 'lane handoff', 'output routing', or 'cross-lane artifact'.
03 SKILL.md 正文
Claude Output Lane
Use this skill to keep Claude output separated by type instead of mixing plans, findings, ideas, drafts, and review notes in one transcript.
Master Lane
The persistent master lane is:
_SYSTEM/reports/claude-output-lane/
This is a report/output lane, not source truth and not runtime state.
Claude output remains advisory until Codex/main verifies and integrates it.
Sublanes
| Sublane | Use for | Typical authority |
|---|---|---|
ideas/ |
Brainstorms, options, design alternatives, rough concepts | proposal_only |
plans/ |
Ordered implementation plans, phase maps, task packets | proposal_only |
findings/ |
Bugs, risks, audit observations, contradictions | advisory_finding |
draft-artifacts/ |
Draft reports, specs, docs, PR text, summaries | proposal_only |
diff-proposals/ |
Unified diffs or patch sketches not yet applied | proposal_only |
reviews/ |
Pressure tests, code reviews, plan reviews, second opinions | advisory_review |
questions/ |
Open questions for owner, Codex, Prime, or Quantum | open_item |
decisions/ |
Proposed decisions requiring owner/Codex acceptance | proposed_decision |
evidence/ |
Bounded command output, screenshots notes, verification snippets | evidence_candidate |
raw-captures/ |
Sanitized transcript excerpts when category is unclear | raw_advisory |
Do not put everything in raw-captures/. Use it only as a fallback.
File Naming
Use:
YYYY-MM-DD_<task-slug>_<lane>_<sublane>.md
Examples:
2026-05-28_search-adapter_quantum_plan.md
2026-05-28_search-adapter_prime_review.md
2026-05-28_plugin-bridge_quantum_ideas.md
Packet Contract
When Claude may write into this lane, the packet must include:
CLAUDE_OUTPUT_LANE_ACTIVE
OUTPUT_SUBLANE=<ideas|plans|findings|draft-artifacts|diff-proposals|reviews|questions|decisions|evidence|raw-captures>
DRAFT_ARTIFACT_ALLOWED path=_SYSTEM/reports/claude-output-lane/<sublane>/<file>.md authority=proposal_only
If the packet does not include DRAFT_ARTIFACT_ALLOWED, Claude should return the categorized output in the TUI response instead of writing a file.
Category Header
Every durable output file should start with:
# <Title>
source_lane: Claude/<Sonnet|Opus>
private_lane: <Quantum Rick|Rick Prime|none>
output_sublane: <sublane>
authority: proposal_only
task: <task>
date: YYYY-MM-DD
codex_status: pending_review
Mutation Boundary
Writing to the Claude output lane is not the same as source mutation.
Allowed with the proper packet:
- draft reports
- plan files
- review files
- idea files
- findings files
- diff proposals
Still requires explicit separate authorization:
- source code edits
- YURI core edits
- deploy config edits
- credentialed tool use
- live service calls
- GitHub/cloud/app connector actions
- commits or pushes
Codex Review Flow
Codex/main should consume the lane by category:
- Read
findings/before summaries. - Read
reviews/before accepting plans. - Read
plans/before implementation. - Read
diff-proposals/only as proposed patches. - Promote
draft-artifacts/only after verification or rewrite. - Move final accepted truth into the task's canonical artifact path, not this lane.
Failure Mode
If Claude writes the wrong category, Codex/main should leave the original intact and create or request a corrected categorized output. Do not silently merge unrelated output types.
Session Notes
2026-07-06
- session: 58m | peak ctx: 0% | compacts: 0
- tools: Read×263, Grep×94, Shell×92, Write×43, MCP:search_contacts×4, MCP:list_chats×2, MCP:open_resource×1, MCP:list_messages×1
- corrections: none
- errors: none
2026-06-03
- session: 1307m | peak ctx: 0% | compacts: 0
- tools: Edit×462, Read×351, Bash×259, WebFetch×99, WebSearch×92, StructuredOutput×29, ToolSearch×14, Workflow×5, Write×5, AskUserQuestion×1
- corrections: none
- errors: none
2026-06-02
- session: 61m | peak ctx: 0% | compacts: 0
- tools: Edit×263, Read×218, Bash×102, WebSearch×51, WebFetch×36, StructuredOutput×12, ToolSearch×7, Write×5, Workflow×2
- corrections: none
- errors: none
04 同仓库更多 Skills
zoom-out
Tell the agent to zoom out and give broader context or a higher-level perspective. Use when you're unfamiliar with a section of code or need to understand how it fits into the bigger picture.
yuri-shura
6-perspective adversarial review for high-stakes turns (architecture decisions, refactor planning, deployment review) — fans out architect, adversary (7-vector attack), maintainer, ops, product, and security perspectives in parallel, then consolidates. Use when user says 'review this decision', '/shura', 'adversarial review', 'what could go wrong', 'architecture review', or 'deployment review'.
yuri-sales-intelligence
Unified sales and outreach skill generating cold outreach openers, social selling sequences, objection rebuttals, and ad creative for any paid platform — multi-persona aware, adapting tone and offer frame to buyer type. Use when user says 'write a cold email', 'outreach sequence', 'objection handling', 'ad copy', or 'sales pitch'.
yuri-code-intelligence
Unified code quality intelligence layer that diagnoses code smells (Long Method, Feature Envy, design pattern issues), prescribes targeted Fowler refactoring techniques, and reviews code for security, performance, and clarity. Use when user says 'review this code', 'refactor this', 'what's wrong with this method', 'code smells', or 'clean this up'.
yuri-adw
Invoke when user asks to run the ADW loop / agentic engineering loop / software factory workflow.
writing-skills
Use when creating new skills, editing existing skills, or verifying skills work before deployment