release
Skill0 starsChanged 8 days agoMITView on GitHub
Release a repository or selected services to production, from version and changelog through main sync, GitHub release, package distribution, deployment and hypercare. Supports --local when CI quota is exhausted; use --dry-run for a read-only release plan.
193 other projects use a version of this.
Adopted by
Most recently taken up by
- yicheng47/runnertheir own version, first seen Nov 13, 2026
- eugene1g/agent-safehousetheir own version, first seen Nov 12, 2026
- BlockRunAI/ClawRoutertheir own version, first seen Nov 11, 2026
- cubeplexai/cubeplextheir own version, first seen Nov 10, 2026
- bex-co/bextheir own version, first seen Nov 10, 2026
- Yeachan-Heo/oh-my-claudecodetheir own version, first seen Nov 9, 2026
Used with
Projects with it also use create-agent-adapter, changelog-style 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.
MicrosoftSkillcreate-agent-adapter
Technical guide for creating a new Paperclip agent adapter. Use when building a new adapter package, adding support for a new AI coding tool (e.g. a new CLI agent, API-based agent, or custom process), or when modifying the adapter system. Covers the required interfaces, module structure, registration points, and conventions derived from the existing claude-local and codex-local adapters.
Also in 5 projects that use this
GrinvSkillchangelog-style
Turn a commit range into CHANGELOG.md entries following common-changelog style. Use when finishing a PR or cutting a release, to write/fix the [Unreleased] section's entries — moving [Unreleased] under a new dated version heading is a separate, automated step (see the release skill / scripts/sync-version.ts), not this skill's job.
Also in 4 projects that use this
GrinvSkilldocs-consistency-check
Check README/manifest.json/server.json/CHANGELOG.md/AGENTS.md and docs/*.md for drift against the actual registered tools and source. Use after adding, renaming, or removing a tool, or as part of a live-audit pass.
Also in 4 projects that use this
Microsoft AzureSkillcomponent-release
Prepares a verifiable Agent Landing Zone UI release. Use for release branches, VERSION, changelog entries, tags, release notes, and compatibility evidence; never publish without explicit human approval.
Also in 4 projects that use this
Microsoft AzureSkilldocumentation-consistency
Keeps AI Landing Zone contributor, user, and operator documentation aligned with shipped Bicep behavior. Use for parameters, defaults, outputs, topology, deployment flow, operations, or releases.
Also in 8 projects that use this
rivet-devSkilllaunch-graphics
Generate agentOS marketing/launch graphics (launch & social heroes, code-snippet cards) from the committed generators in docs-internal/graphics. Use when asked to make, update, or render an agentOS launch image, social image, "agentOS Apps" graphic, or technical code-snippet image.
Also in 3 projects that use this
- +3 more from Grinv
Other versions193 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 193
- paperclipai/paperclip★ 114ksame name
- modelcontextprotocol/servers★ 92ksame name
- remotion-dev/remotion★ 73ksame name
- traefik/traefik★ 66ksame name
- usememos/memos★ 64ksame name
- tw93/Pake★ 64ksame name
- amruthpillai/reactive-resume★ 46ksame name
- reactive-resume/reactive-resume★ 46ksame name
- AmruthPillai/Reactive-Resume★ 45ksame name
- Yeachan-Heo/oh-my-claudecode★ 41ksame name
183 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 repository27 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- agent-first-docsSkill
- analytics-reviewSkill
- autopilotSkill
- browseSkill
- create-prSkill
- design-uiSkill
- foremanSkill
- harness-auditSkill
- image-genSkill
- implementSkill
- investigateSkill
- marketing-videoSkill
- plan-draftSkill
- qaSkill
- reconSkill
- release-prdSkill
- review-prSkill
- semantic-decisionSkill
- setup-projectSkill
- teamSkill
- telemetry-reviewSkill
- testSkill
- token-auditSkill
- triageSkill
- ui-auditSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 6 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 183 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 11
Follow preamble for bootstrap/telemetry and the Auto-Decision Framework for decisions.
- Very specific: line 38
Do not expand a service release to siblings or upgrade unrelated local tools.
- Very specific: line 47
Read applicable AGENTS.md/CLAUDE.md, the profile explicitly linked there, otherwise .babysit/release.md or RELEASE.md, git policy, deploy …
- Distinctive: line 52
Use profile contract when a profile is missing or incomplete.
- Very specific: line 93
Before mutating, write a release record under <git-common-dir>/bbs-release/<tag>.md (resolve with git rev-parse --git-common-dir), or the …
- Very specific: line 99
Use pending / running / verified / failed / N/A.
- Distinctive: line 115
Follow repo merge/rebase policy without force pushes.
- Distinctive: line 120
Version and changelog: Update the authoritative version and its required mirrors, lockfile metadata and generated release files together.
- Very specific: line 128
Verify the candidate: Run the profile's required checks and meaningful QA, including a non-happy-path case.
- Distinctive: line 133
Commit intended changes according to repo policy, then fetch/reconcile main again before pushing.
- Distinctive: line 168
Post-deploy and hypercare: Verify public health/readiness plus changed user behavior and a non-happy-path case against production.
“Common” is measured: kept by at least 6 in 10 of the 183 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
release
Own the release through verified distribution, production behavior and local consumer updates. Builder owns this delivery skill; autopilot's QA handoff does not invoke it or grant release permission.
Follow preamble for bootstrap/telemetry and the
Auto-Decision Framework for decisions.
Shared refs are filesystem paths beside this skill's directory, so read them by path, not as skill://.
For --dry-run, skip the mutating bootstrap and telemetry; read context directly.
Invocation and scope
Options are agent instructions, not a bbs release CLI command:
release [project/service] [version|patch|minor|major]: execute the requested release using the repo's normal pipeline and delivery policy.--local: run all checks, builds and distribution from this machine; bypass automatic CI. Read local mode before any remote mutation. GitHub/registry APIs and production rollout remain remote.--dry-run: inspect local files and remote state, report resolved stages, commands, versions and gaps. No edits, commits, pushes, tags, workflow state changes, build/test hooks, publication, deployment or installs.--prepare: perform local version/changelog edits and checks; stop before remote mutations. Combine with--localto prepare that delivery path.--resume <tag>: inspect the recorded release and real external state; continue incomplete stages for the same version, SHA and mode.
A request to execute a production release authorizes its selected targets and
their required delivery steps; do not repeatedly ask for confirmation. Creating
this skill/profile, a dry run, a preparation request, or having credentials
does not authorize executing a release. Honor explicit bounds and repo policy;
escalate only unresolved scope, permission or irreversible-data decisions.
Do not expand a service release to siblings or upgrade unrelated local tools.
A Team phase-release assignment is such a request: it names the phase, the
delivered source SHA and its go (the approved project design for an auto
phase, the resolved release approval for an ask phase). Release exactly
that SHA's base, and end with the Output block so the lead can record the
verdict and record path with bbs team release --file.
Resolve the project contract
Read applicable AGENTS.md/CLAUDE.md, the profile explicitly linked there,
otherwise .babysit/release.md or RELEASE.md, git policy, deploy runbook,
version sources and actual release workflows/scripts. Resolve remote from git;
never infer a submodule's release identity from its parent's remote.
Use profile contract when a profile is missing
or incomplete. An inline section in AGENTS.md is equally valid. Derive facts
from code/CI first; do not require a new config file to release a known project.
Missing facts block only their dependent actions. Every stage needs an owner
(agent or a named workflow), concrete verification, or N/A with a reason.
In local mode the agent owns every required stage.
Discover CI and assign execution
Scan the target repo's .github/workflows/ YAML files at the candidate and
remote base, plus referenced local reusable workflows, composite actions and
release/build scripts. Follow job needs, conditions, matrix targets, working
directories, environment requirements, artifact handoffs and lifecycle hooks.
Inspect referenced remote action/workflow definitions when their release work
has no known local equivalent. Do not assume a shell run step is portable to
this machine or that the profile's workflow list is still complete.
Persist an execution map in the release record before delivery:
stage | workflow/job or script | normal executor | local command + cwd |
dependencies | success evidence. Derive this from the profile and actual
CI; report conflicts. In normal mode, trigger and wait only for stages owned
by CI, while the agent executes its assigned stages. Match the exact workflow,
source SHA and intended run; wait for its required jobs to finish successfully,
then verify published outputs. Poll with progress updates; failed, cancelled,
quota-blocked or missing required runs block dependent stages.
--local overrides every CI executor with its mapped local equivalent and
preserves dependencies, checks, platforms and artifact verification. No CI
watch/wait or CI artifact dependency remains. Inspecting run state to detect
a competing release is allowed; it is not a delivery gate. A missing local
equivalent is a blocker, never a skipped job or silent fallback to CI.
Resolve branch, selected components, version/tag policy, changelog baseline, checks, artifacts/registries, deploy order, rollback target, hypercare window, failure thresholds and local CLI/consumer updates. If version was not given, use the repo rule; absent one, recommend the smallest semver bump supported by the API/behavior changes and explain it. Do not overwrite an already prepared version or use a prerelease as stable. Preserve independently versioned packages.
Release record and recovery
Before mutating, write a release record under
<git-common-dir>/bbs-release/<tag>.md (resolve with git rev-parse
--git-common-dir), or the caller's existing evidence destination. It is local
git metadata, never a tracked secret-bearing report. Record scope, profile,
mode, old/new version, baseline, source SHA, artifact hashes, permissions,
workflow restoration state, rollback identity, timestamps and stage results.
Use pending / running / verified / failed / N/A; record intent before an
external action and evidence after it. Never store credentials or raw secrets.
On resume, the record is a pointer, not proof: recheck remote tag SHA, release draft/public state, published artifact integrity and deployed revision before skipping work. A timeout is not proof publication failed. Query before retrying; retry a transient read at most three times, and retry a mutation only after confirming its outcome. Do not move published tags or replace published assets; a different artifact/source needs a new version. Report partial success rather than undoing a package publication or declaring the whole release complete.
Flow
- Inspect and reconcile. Inventory dirty/staged files, branch, local commits, remote main, tags, authentication and concurrent release work. Fetch the configured remote/base and tags; inspect ahead/behind/divergence. Follow repo merge/rebase policy without force pushes. Preserve unrelated user changes; use a clean release checkout when the build context would include them, without creating a new branch contrary to repo policy. Do not stash/reset another person's work. Include only requested changes.
- Version and changelog. Update the authoritative version and its required mirrors, lockfile metadata and generated release files together. Derive changelog from actual changes since the previous release of this component (diff + commits, not just commit titles). On the first release, state the baseline rather than guessing a previous tag. Describe user-visible changes, breaking changes, migration/compatibility notes and upgrade steps. Reuse this content for GitHub notes; exclude secrets/private operational details.
- Verify the candidate. Run the profile's required checks and meaningful
QA, including a non-happy-path case. Library/CLI QA can be a failing input
or isolated install/import; UI QA uses
browse/qawhen available. In local mode run CI-equivalent checks locally, including all distributed targets; unavailable required checks block release rather than becoming a waiver. Commit intended changes according to repo policy, then fetch/reconcile main again before pushing. A changed candidate invalidates affected verification and artifacts. Record the final release SHA; remote main must contain it before release. For a PR policy, require the authorized landing path and check the resulting merge SHA; never silently bypass branch protection. - Sync remote main and release on GitHub. In normal mode, let the named
pipeline own tagging/distribution it already implements; follow the run for
this SHA (
gh run list --commit <sha> --workflow <file>), not just the last green run. CI quota exhaustion is a blocker unless--localwas requested; offer that concrete recovery path. For agent-owned releases, push main, create/push the exact configured tag at the verified SHA and create the release with--verify-tag --notes-file. Verify the remote tag resolves to that SHA. Stage a draft when artifacts/distribution gates must precede public release. Respect the project's publication order; a pipeline that publishes npm before making its GitHub draft public is valid. Before any push/tag/release event in local mode, apply the workflow guard. - Distribute. Publish only selected packages/binaries using the profile's registry, access, channel and credentials. Build from the verified source; inspect the actual packed artifact and publish those exact bytes. Read npm distribution for npm targets. Verify download, exact version, integrity, expected assets and channel/formula pointers; a green publish command alone is insufficient. If CI owns this, verify its actual outputs. Complete/publish the GitHub draft after its required gates.
- Deploy. Capture the previous active revision/image first. Deploy only
selected apps in dependency order from the same source/artifacts; record
image digests and production origin. Never substitute an ambiguous
latestimage. Follow the current runbook for migrations and traffic changes; image rollback does not roll back schema. In local mode build locally and push the resulting image, then call the rollout tool directly or a proven--skip-buildwrapper with the exact tag. A wrapper invoking Cloud Build is not a local build. No deployed app meansN/A, not an invented deployment. - Post-deploy and hypercare. Verify public health/readiness plus changed
user behavior and a non-happy-path case against production. Use safe test
accounts; do not create charges or messages without scope authorization.
Compare errors, latency, restarts and relevant queues/jobs to the pre-deploy
baseline for the configured window; record timestamped observations at
start and end plus intermediate samples. No observable traffic/metrics
means a named evidence gap, not a clean bill of health. Apply the configured
stop/rollback rule when breached, verify recovery, and stop later dependent
deploys. Otherwise update the selected local CLI/consumer from the released
artifact, inspect PATH collisions and run version + smoke checks. Pin the
version, or prove the upgrade channel resolves to this release. Verify any
website/download catalog the release is expected to update. Restore local
mode's workflow states on success, failure or interruption; missing
restoration blocks completion. For the 2found tools, a release also
publishes agent-facing docs:
2found.devserves a committed snapshot (content/<product>/source/+snapshot.json+catalog.json), and the start guide links those pages for the agent to read. After the tool release lands, sync that snapshot to the released revision and release2found.dev— then verify the updated pages actually render. A release that leaves the docs pointing at the old source SHA ships stale instructions to new agents; record it as a concern. Finish only after required post-release work.
Output
STATUS: DONE | DONE_WITH_CONCERNS | NEEDS_CONTEXT | BLOCKED
VERDICT: RELEASED | PREPARED | DRY_RUN | PARTIAL | ROLLED_BACK
RELEASE: <repo + component + version/tag + source SHA + mode>
STAGES: <version/changelog; main; GitHub; distribution; deploy; hypercare; local update>
EVIDENCE: <record path + release URL + artifact/deployed identity + checks>
WORKFLOWS: <unchanged or restored IDs; outstanding restoration if any>
NEXT: <none or exact incomplete action>RELEASED requires every applicable stage verified. PREPARED and DRY_RUN
mean their requested scope completed, never a production release. Required
publication/deploy/observation/update failures are BLOCKED + PARTIAL;
DONE_WITH_CONCERNS is for nonblocking concerns only.
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.

