ZenUml/conf-app / .claude/skills/auto-lane
auto-lane
VerifiedDecide whether an agent session may merge a PR on ZenUml/conf-app without asking (non-destructive rule), and operate the daily auto-lane merge job (classify a diff, dry run, backtest, kill switch). Use when a PR is green and the next step is "merge or ask", when asked "can this be auto-merged", "why did the auto lane skip PR N", "run the auto-lane dry run", "自动合并", "非破坏性", or before proposing any automation that merges.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.6 KB |
| scripts/classify-pr.shexecutable | script | 2.1 KB |
SKILL.md
View on GitHub---
name: auto-lane
description: Decide whether an agent session may merge a PR on ZenUml/conf-app without asking (non-destructive rule), and operate the daily auto-lane merge job (classify a diff, dry run, backtest, kill switch). Use when a PR is green and the next step is "merge or ask", when asked "can this be auto-merged", "why did the auto lane skip PR N", "run the auto-lane dry run", "自动合并", "非破坏性", or before proposing any automation that merges.
---
# Auto lane
Rules and their reasons: `docs/policies/auto-lane.md`. Read its "Agent sessions" section before the
first merge of a session. This file holds the commands only.
## Input and output
| Input | Output |
|---|---|
| A PR number | Merge without asking, or offer `[AI] Merge PR #N` with the reason |
| A diff (`base`, `head`) | The daily job's lane and level for it |
| "Why was PR N skipped?" | The job's skip reason from a dry run |
## 1. May the session merge this PR?
```bash
.claude/skills/auto-lane/scripts/classify-pr.sh <PR> # reads only; needs gh logged in
```
| Result | Action |
|---|---|
| exit 0 | Non-destructive by path, preconditions met: merge (step 2). |
| exit 1, `blockers` not empty | Do not merge. Fix the blocker (CI, draft, conflict) or report it. `mergeable:UNKNOWN` is transient: run again. `runs-pending:<workflow>`: a run is still waiting (often `PR validation` on the staging lock) and no check lists it yet; wait for it. `required-check:<name>`: that check has not passed on this head. |
| exit 1, `readDiffFor` not empty | Read those files' diff against the destructive table in the policy. Non-destructive: merge. Destructive: offer `[AI] Merge PR #N` and wait. |
| exit 2 | Fetch or classifier error. Do not merge. |
After a session merge, report it in the same turn: PR link, merge commit, one-line risk reading,
and what is not yet verified. Do not ask before a non-destructive merge and do not offer it as an option.
A Draft the session finished itself is marked Ready first (`gh pr ready`); see the policy.
`jobLevel` is what the daily job would do. It does not bind a session.
The script cannot see what a diff does. A change outside `readDiffFor` that deletes stored data or
publishes a release is still destructive: the path list is a prompt to read, not the rule.
## 2. Merge
Use the `land-pr` skill. Pass `--match-head-commit <head from step 1>` and a merge commit. Then
watch `Build, Test and Draft Release` on `main`. Do not release to production.
## 3. Daily job
```bash
node scripts/ci/auto-lane-merge.mjs --dry-run # decisions for every open Dependabot PR
gh workflow run auto-lane-merge.yml # same, on GitHub (dry run by default)
node scripts/ci/auto-lane-policy.mjs --base <sha> --head <sha> --json # lane and level of one diff
gh variable get AUTO_LANE_ENABLED # kill switch; anything but "true" is off
```
Setting `AUTO_LANE_ENABLED` changes what merges unattended: do it only on the owner's explicit pick.
## 4. Backtest after a classifier change
For each merged PR in the window, classify `<merge>^1..<merge>^2` and count lanes:
```bash
gh pr list --state merged --base main --limit 300 --json number,mergeCommit \
--jq '.[]|[.number,.mergeCommit.oid]|@tsv' |
while IFS=$'\t' read -r n sha; do
node scripts/ci/auto-lane-policy.mjs --base "$sha^1" --head "$sha^2" --json 2>/dev/null |
jq -c --arg n "$n" '{pr:$n,level,lane}'
done | jq -s 'group_by(.level)|map({level:.[0].level,count:length})'
```
Update the backtest table in the policy when the counts change. A revert commit that classifies
above `report-only` is a false approval: fix the classifier first.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |