merge
Skill1 starsChanged 17 days agoApache-2.0View on GitHub
Land a reviewed cland task through the hierarchy verdict gate, or handle a gate refusal.
58 other projects use a version of this.
Adopted by
Most recently taken up by
- onetimesecret/onetimesecrettheir own version, first seen Nov 16, 2026
- Trixx4191/asrartheir own version, first seen Nov 15, 2026
- automagik-si/genietheir own version, first seen Nov 15, 2026
- ryanportfolio/Corewise.Academytheir own version, first seen Nov 12, 2026
- ToolJet/ToolJettheir own version, first seen Nov 12, 2026
- toy-crane/claude-code-playbook-templatetheir own version, first seen Nov 10, 2026
Used with
Projects with it also use addskill, apply-firmware and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ryanportfolioSkilladdskill
Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.
Also in 6 projects that use this
ryanportfolioSkillapply-firmware
Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.
Also in 6 projects that use this
- +10 more from ryanportfolio
Other versions58 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 58
- microsoft/vscode★ 195ksame name
- remotion-dev/remotion★ 73ksame name
- withastro/astro★ 64ksame name
- ToolJet/ToolJet★ 41ksame name
- Openpanel-dev/openpanel★ 7.3ksame name
- raine/workmux★ 3ksame name
- onetimesecret/onetimesecret★ 3ksame name
- jtenniswood/espcontrol★ 1.2ksame name
- K9i-0/ccpocket★ 1.1ksame name
- r1n7aro/Locus★ 759same name
48 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository21 other files it gives its agents.
- AGENTS.mdAGENTS.md
- codebase-designSkill
- code-reviewSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- grillingSkill
- handoverSkill
- implementSkill
- onboardingSkill
- plan-epicSkill
- prototypeSkill
- qaSkill
- releaseSkill
- researchSkill
- retroSkill
- tddSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- wayfinderSkill
- writing-for-agentsSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 1 is distinctive, and 2 are very specific to this project.
0
Common
Most of the 52 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Use the hub's merge tool with item: T-<n>, or cland merge T-<n> from the supervising Lead's or Coordinator's shell.
- Very specific: line 65
If merge-recovery names pending cland sync, finish that sync retry before merging.
- Distinctive: line 87
Read the report-only assessment on the item as path/dependency evidence, never as a Reviewer verdict.
“Common” is measured: kept by at least 6 in 10 of the 52 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Lane order uses follows between epics, separate from blocked_by. A follow is met when its predecessor closes, or has task children all closed with its last ticket merged. When every ticket of the predecessor’s last stage is in review, cut the following epic’s tickets with to-tickets and write its plan-epic; launch the first stage on the last merge once blockers are closed and follows met. The predecessor’s QA demo runs in parallel. Until either demo verdict is recorded or the predecessor closes, the merge gate holds follower tickets whose diff overlaps files changed by that predecessor’s merges on dev; tickets with no overlap can merge.
Enqueue also posts advisory merge-demo-warning notices in the candidate's topic (Agents fallback), to the Coordinator and item lead, naming overlapping epic/files regardless of follows. An open epic is under demo with an active unexpired lease, or all task children closed and last ticket landed on dev without a hub QA demo verdict; stage QA is separate. Closed epics and recorded demos without active leases are excluded. Unreadable lease/history evidence posts merge-demo-warning-unavailable; enqueue succeeds while authorization, lane and freshness refusals remain enforced. Inspect warnings when sequencing demo and landing. See hub/README.md for the contract.
Use the hub's merge tool with item: T-<n>, or cland merge T-<n> from the supervising Lead's or Coordinator's shell. Task work reaches the task base (dev by default) only through this gate.
A merge-bare refusal names core.bare=true in the project root (not the intentionally bare territory). Report it to the responsible item lead. Inspect the shared Git config before the repository owner restores it with git -C '<root>' config --local core.bare false; the hub never edits it automatically. Reproduce Git environment leaks only in disposable repositories with sealed HOME/XDG/Git environments.
The supervising Lead merges scheduled tasks within its claimed subtree; the Coordinator merges unscheduled tasks and tasks outside a Lead's scope. Perform the merge yourself. Never delegate a merge to a sub-agent: it inherits the launching session's authority.
On the hub’s ready status event, read the item with task.view and call the gate yourself. This one event starts the supervising Lead’s or Coordinator’s turn and stands for the item lead’s ready report. The claimant leads its review and QA loop: it requests the Reviewer verdict and, when task.view’s qaPolicy has a non-none scope and when: per-item, the QA verdict, and fixes not-clear findings before requesting a new pass. Reports address lead T-<n>; the nearest container Lead, then Coordinator, is the fallback for an unclaimed item. Let the lead handle the loop instead of relaying verdicts. Only the latest hub-written verdict of each kind counts: the writer must have held the corresponding role at write time, must not claim the item, and the verdict must be clear and current for the task branch. A Coordinator's rescue-run verdict does not satisfy the gate.
Default subscriptions give each Lead its subtree outside descendant leads and the Coordinator the remainder, plus the immediate ready exception. Other events are digested at the next turn; use subscribe for additional subjects and subscriptions to inspect them instead of polling. Send routine status updates as kind progress; blockers and actionable refusals use question or report.
Request source review first. Only after a clear source verdict does the Implementer run cland verify (the transition alias is cland suite) at the exact reviewed head. A fix or merge that changes the head needs a new source review before a new run. An unchanged head may reuse an earlier green run through the tree-hash cache. Require the same-head successful result and the Reviewer’s final verdict before merging; queueing or a source verdict alone is not a pass. The gate still relies on current hub-written review/QA verdicts. See docs/agents/tests.md.
With [merge] queue = true, a task verify runs T0–T2 plus changed T3/T4 files and tiers.json mapped T4 read-path guards once (duration procedure: docs/agents/tests.md) when both the candidate and hosting hub enable it. The hub supplies T3 in the merge batch. At stage close, epic close or before release, the Coordinator runs cland verify all --occasion stage|epic|release --report <topic> on clean root dev at fetched origin/dev. QA may set epic for its integrated demo. These occasions include T4 and may shard across remote slots; merge batches use one slot. The nightly soak includes T4 on one droplet. Read docs/agents/tests.md for selection, spend, stage-platform evidence and cache rules.
Periodic dev T3 runs only when the verification candidate tree differs from the last started run's tree; tracker-only commits retain that tree. Eligibility uses the shared exclusions in hub/src/verification-tree.ts. See docs/agents/tests.md, "Periodic dev T2/T3 and release gate".
After a green stage cover, the Coordinator adds missing duration baselines with node hub/test/durations.ts refresh --add-only <durable-log-path> on dev, commits and publishes them. Use add-only until T-931 decides allowance refresh; see docs/agents/tests.md. With a tier manifest, both refresh modes skip and print identities outside its owned files, so imported site/test cases need no manual removal; invalid existing keys still refuse shard planning. Test renames and moves remove stale duration keys in their own item: both admission paths refuse unmatched identities. Before an epic's QA demo, re-own the hub/test/durations.json allowances it owns to an open release or category, keeping added, expires and reasons: epic close refuses and names each remaining key, even with --override, and a later re-owning commit leaves the demo verdict stale.
A green same-head result from either executor satisfies the gate's suite evidence. A Linux droplet result lists macOS-only cases as skips; those cases run in the release's final local run ("Release v0.1.2 from the item"). Read the executor and platform in the result, and the droplet's lifetime and cost; a Linux pass leaves the final macOS checks to that release run.
Call merge when that evidence is present. For queue-enabled code changes the CLI returns T-<n>: queued <n>, the current FIFO position, rather than a landed commit. The entry pins the reviewed branch/head and requester. Read cland merge queue [--json] (MCP merge.queue) for entries and the current batch; cland verify queue also shows its members. To withdraw an entry, use cland merge T-<n> --cancel (MCP merge with cancel: true), under the same role/subtree permission as merge. It returns cancelled or not queued. A changed/deleted head, closed item or not-clear verdict drops the entry. Let an active worker finish before candidate checkout cleanup; cancellation does not grant permission to stop someone else's verification.
The hub takes the oldest [merge] batch entries (default 3) when full or after batch_wait (default "10m"). It builds squash candidates in order, checks immutable lint and duration budgets, then runs T0 (check) before the configured batch_tiers on the detached candidate, adding Web after T1 when its diff touches web/ (after configured tiers if T1 is absent). New T3 identities repeat five times against the last green batch or stage receipt. On green it lands the members in order. Tracker-only dev movement reuses the tested tree; a changed candidate tree rebuilds and reruns. A red batch bisects failing files or the failed T0/Web step: leave each member out against the pinned base, then reject a unique culprit or repeat the original failures once. A green step diagnosis or file-only recovery of a Web-containing plan requires the original pinned full plan on the selected candidate before landing; a completed green file-only repeat can land directly only when the original plan has no Web step. Unresolved failures reject all members. Inconclusive original runs or diagnostics hold the batch for the Coordinator. The hub retains each run, tree, log, duration and cost; diagnostic jobs supply no full-suite cache evidence. See docs/agents/tests.md#queued-merge-batches for T0 receipts and the combined-tree case.
For a recoverable landing block, resolve the named cause and let the queue retry automatically with its retained green evidence; see docs/agents/tests.md#queued-merge-batches for the codes. Other held failures require human or Coordinator action as the notice states.
When the item is closed with a durable merge receipt, a failed follow-up is reported for that item and the batch continues. Repair the named step with cland merge T-n; the landing is not repeated. A retained blocked batch whose members are all landed or completed clears on the next tick, including startup, after admitted workers finish. Follow-up notices have a durable delivery obligation separate from the batch. A failed notice retains the original follow-up reason and separate report error; later ticks retry it even after candidate cleanup or hub restart, without holding remaining landings or the next batch.
Results reach Agents, the requester and each item lead. The supervising Lead or Coordinator handles queue refusals below; the item's Implementer owns any fix and new review/verify cycle. When the queue is disabled, or changes stay wholly in the documentation/tracker allowlist below, the gate lands immediately with its required verdicts. The allowlist does not include skills or role configuration.
Lockfile changes or manifest changes to dependencies, devDependencies,
optionalDependencies, peerDependencies, overrides or workspaces add a
required root npm install notice to queue entries, landing results and the
durable landing comment. Scripts/version/metadata-only changes omit it; missing
or malformed manifests retain it conservatively. Batch landing
sends it to the enqueuer and item lead. Stop shared readers before updating the
root install. The merge gate never runs npm or replaces shared node_modules.
Landing keeps the task branch unchanged, creates one squash commit, closes the item and topic, releases the claim and removes its worktree and branch. Removal waits while a session stands inside; stop finished sessions so cleanup can complete.
On refusal:
forbiddennaming a level or subtree: report the request tolead T-<n>for your claimed node. The Coordinator keeps releases, sync, push, unscheduled merges and Lead launches.- Missing, invalid, not-clear or stale verdict: send the refusal to
lead T-<n>with kindreport; the lead requests the new verdict from the role named by the refusal. For an unclaimed item, request it yourself. verdict-missing: preserve the durable record and restore its missing or altered comment; an older clear cannot replace it.- Freshness pending: let the async cache settle, then retry once; if still pending, report the refusal.
merge-bisected: the sole member of a red batch, or the unique member whose removal made the failing files green, was rejected. Send the failed identities and retained receipts tolead T-<n>for diagnosis and a fix. The Implementer obtains fresh source review and task verification; the supervising Lead or Coordinator enqueues the repaired head.merge-bisect-unresolved: every member was rejected because no unique removal was green and the original failing files did not supply complete green proof on repeat. The Coordinator diagnoses from the original and file-job receipts and decides which items need fixes or may re-enqueue. Item leads preserve their heads while awaiting that decision.merge-conflict: send the named conflict or branch to the item's Implementer to integrate dev in its worktree. A changed head needs source review and task verification before the supervising Lead or Coordinator enqueues it again.merge-budget: the immutable candidate could not be inspected or exceeded an enforced tier projection or T3 growth limit. The Implementer resolves the named inspection failure, reduces growth or adds a valid reasoned allowance underdocs/agents/tests.md, then obtains fresh review and verification before retry. It does not bypass the budget. The Coordinator owns the report-to-enforce rollout.merge-hold: wait for the named predecessor’s demo verdict or closure, then retry the same head. Keep thefollowsrelation and the gate in place.merge-unpublished: the root’sdevhas commits absent from the territory; a Lead reports it tolead T-<n>for its own node. The Coordinator commits remaining tracker work, runscland pushfrom its own shell, then retries merge. A refused merge leaves refs and checkout unchanged and names those unpublished commits.- Touched root edits or wrong checkout: preserve the root's work and restore the task-base checkout before retrying.
merge-recovery: neither integration ref landed the prepared commit, but checkout restoration is pending. Preserve the durable receipt, remove the named obstruction, and retry the same item; it restores the checkout before rechecking the gate.- If
merge-recoverynames pendingcland sync, finish that sync retry before merging; its receipt pins dev until recovery completes.
Only the human's own shell may run cland merge T-<n> --override "<reason>", for changes confined to .cland/tasks/**, .cland/RULES.md, docs/**, GLOSSARY.md, files named README.md in any directory, and qa/briefs/**. The shared tool schema recognizes override and the spoke preserves it, so a session with merge access receives the human-only forbidden refusal; recognizing the field grants no authority. Unknown fields still return invalid_argument. The gate checks the immutable merge candidate against its task base, including deletions and both sides of renames. merge-override-scope names every other changed path; send that work through review. Configuration files and skill instructions require review even when they are Markdown. When a permitted override decision is needed, ask the human through the Coordinator's ask; the reason goes on the item and into the journal. Override skips verdicts and criteria; conflicts and Git safety still apply.
Report the returned commit and cleanup status in the item's topic. If cleanup keeps uncommitted work or a branch moved since merge, preserve its branch and worktree. Use the human's cland worktree prune only after the work is safe to remove.
A rollback retains an exact inverse-patch receipt; re-landing the work is a separate item. Attribute the receipt under rule 17 (A-8634 R13; T-970).
If the result says merged, follow-up failed: <step>, the returned commit has already landed. Repair the named rollback, checkout, publish, close, release, topic or cleanup failure and retry merge for the same item. Checkout recovery preserves staged and untracked work; reconcile any changed index before retrying it. The hub verifies its durable prepared landing against Git and completes the remaining steps without creating another commit, including after a hub restart.
Queued member conflicts
For queue-enabled code changes, enqueue checks pinned queued heads applied onto
dev in FIFO order. A merge-conflict names the queued item/head and conflict
files; route it to that item's responsible lead to coordinate the overlapping
changes. Batch drop notices for retained queues use the same diagnosis.
Dev conflicts retain the rebase-on-dev instruction. A refusal changes
neither the existing queue nor refs/checkouts; obtain fresh review for any source
fix before verification and retry. No additional role permission is needed.
Merge-delta advisory comments
Read the report-only assessment on the item as path/dependency evidence, never
as a Reviewer verdict. The hosting hub preserves patch-id freshness and existing
gates; the item lead still obtains merge-delta review before verification under
rule 11. patchIdCurrent records carry-over eligibility, not an observed gate
acceptance. The conservative rule and unknown graph/budget outcomes are defined
in hub/README.md, "Merge-delta assessments (report mode)". Enforcement awaits
T-1486's separate replay decision. Existing role clauses cover this flow.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

