Skip to content
Agent Orchestration
Skill

/codex-lane

Operate Claude Lane Stack schema-v2 runs from Codex. Use for lane status, start, resume, review, verification, evidence, or acceptance.

BOOST
From plugin
claude-lane-stack
12464 skills24 agents5 commands12 hooks
+1
Install
$ npx -y skills add VKirill/claude-lane-stack --skill codex-lane --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/codex-lane

Context preview

The summary Claude sees to decide when to auto-load this skill.

Operate Claude Lane Stack schema-v2 runs from Codex. Use for lane status, start, resume, review, verification, evidence, or acceptance.

SKILL.md

codex-lane.SKILL.md
name: codex-lane
description: Operate Claude Lane Stack schema-v2 runs from Codex. Use for lane status, start, resume, review, verification, evidence, or acceptance.

Codex Lane

Use the installed typed wrappers under `~/.agents/bin`. Keep the run directory, task file, and project root supplied by the user or existing run state. Do not invent a run, task id, path, model, or receipt.

Dependencies

Check the wrappers before acting when the installation is uncertain:

command -v run-controller lane-ctl night-review
run-controller --help
lane-ctl --help

The wrappers may be invoked by absolute path (`~/.agents/bin/...`) when `PATH` does not include `~/.agents/bin`. Their lifecycle state lives under the run's `.agents/runs/<slug>/` directory.

Run and resume

For an existing schema-v2 run, validate and start the durable controller. Keep the run's configured writer by default; add `--provider codex` only when Codex is the explicitly selected writer:

run-validate --run-dir "$RUN_DIR" --phase pre-dispatch
run-controller start --run-dir "$RUN_DIR" --project-cwd "$PROJECT_CWD"
run-controller watch --run-dir "$RUN_DIR" --timeout 240 --json

`run-controller start` is idempotent and resumes persisted controller state. Use a fresh bounded `watch` call after an interrupted session. Read the controller receipt with:

run-controller status --run-dir "$RUN_DIR" --json

Preserve the run's selected native profile, provider, model, reasoning effort, and service tier unless the operator explicitly asks for a change. Do not silently route a run to Codex or override those values.

For one typed lane instead of the DAG controller (use `--provider codex` only when explicitly selected):

lane-ctl start --run-dir "$RUN_DIR" --task-file "$TASK_FILE" \
  --project-cwd "$PROJECT_CWD"

Status and evidence

Read compact state first, then inspect the bounded journal or a specific log:

lane-ctl status --run-dir "$RUN_DIR" --task-id "$TASK_ID" --json
lane-ctl events --run-dir "$RUN_DIR" --task-id "$TASK_ID" --json
lane-ctl tail --run-dir "$RUN_DIR" --task-id "$TASK_ID" --source report --lines 80

Require a trusted complete `report.md`, matching `report_sha256`, provider exit 0, and the current attempt before treating a writer as ready for L1. The writer's report is evidence; it is not acceptance.

Review, verify, accept

`lane-ctl` has no `review` subcommand. Run the installed read-only review entrypoint when a review is requested:

night-review "$PROJECT_CWD" --project-cwd "$PROJECT_CWD" --run-slug "$RUN_SLUG"

Run the task's immutable verification and ownership checks:

lane-ctl verify --run-dir "$RUN_DIR" --task-file "$TASK_FILE" \
  --project-cwd "$PROJECT_CWD"
check-owns-paths "$TASK_FILE" --run-scope

When the task/run requires review, produce the task-specific receipt before acceptance. Use the immutable pre-task commit recorded by the run as `BASE_REF`; if it is missing, report the missing baseline. A repository-wide night review does not substitute for this receipt:

night-review-engine review-task --repo "$PROJECT_CWD" --run-dir "$RUN_DIR" \
  --task-file "$TASK_FILE" --base-ref "$BASE_REF" \
  --output "$RUN_DIR/artifacts/$TASK_ID/review.json"

Finally apply the typed acceptance gate:

lane-ctl accept --run-dir "$RUN_DIR" --task-file "$TASK_FILE" \
  --project-cwd "$PROJECT_CWD"

Acceptance must produce `artifacts/<task-id>/acceptance.json` and a current accepted state. If any receipt is stale or missing, stop and report the exact gate failure instead of editing control-plane artifacts.

Do not run provider commands directly, rewrite `report.md`, bypass `lane-ctl`, or claim browser or user-specific verification from a fixture. This plugin does not replace native Codex compaction or add an MCP server.

Read more
Ships withclaude-lane-stack

Claude Lane Stack is a self-hosted toolkit for coordinating coding agents through CLI processes, task contracts, Git workspaces, and verification receipts (bin/run-controller:117-159, schemas/task-v2.schema.json:1-60).

Get the whole plugin

Other skills on claude-lane-stack.