agentleFS
mariomilemariomile/builder-os/skills/release-ops/SKILL.md

release-ops

Skill2 starsChanged 49 days agoMITView on GitHub

Takes a product to production and work to users safely: environments from the first slice, a rollout the person picks, a tested rollback, the baseline before exposure, release notes. Use when deploying, going live, setting up hosting, or releasing.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Prepare and report readiness: Write .builderos/initiatives/{initiative}/06-release.md per the release template, headings, bold labels and …

    line 25

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Release Ops

Phase 6 puts the work in front of users in a way that can be undone, and captures the number that makes phase 7 possible. RELEASE READY means the preparation exists; SHIPPED means exposure was observed.

Mode. Standalone delivers the requested release plan, checklist, rollback design or notes from supplied context, states missing readiness evidence, and deploys only on existing explicit authorization. The lifecycle runs every step and gate 6. Both per operating modes. Setting up production (environments, domain, secrets, email sender, live keys, legal pages, a preview per build) is its own mode, run as phase 5's slice zero or alone: follow production setup. Release notes follow product communication; tracking-standards checks instrumentation; the analytics shapes measure a baseline. Read the release method for strategy, rollback, baseline window and the evidence a claim needs.

Procedure

  1. Read 05-build-plan.md, 04-spec.md, 03-solution-bet.md, 02-definition.md: the verified build, the tracking plan, the kill criteria, the success metric with its phase 2 definition. Done when the build plan exists with gate 5 passed.
  2. Put the rollout to the person as a decision card (flag, internal, canary, percentage, full): the two or three that fit, each with its exposure per step, what is watched between steps, the numeric condition for proceeding, its cost, and a recommendation. Full release fits a reversible change, a low blast radius, or volume too low to slice; say which applies. Done when **Chosen by:** under ## Rollout records the person's answer (gate 6.13).
  3. Design the rollback (gate 6.1): a specific reversing mechanism, a named owner with how to reach them, and a test. Say what happens to data written while the feature was live, even when nothing is written. With a migration, state whether the change is actually reversible; the inverse migration is a candidate, checked against the data. Done when mechanism, owner, data answer and test are written.
  4. Test the rollback once, before exposure, and record what was done and observed. Done when the observation is recorded.
  5. Verify production instrumentation: events arriving from the production path, with the production configuration, before the first user. Done when each tracking-plan event is observed from production.
  6. Capture the baseline (gate 6.2) before exposure: the phase 2 success metric by its phase 2 definition, over a window covering the product's natural rhythm, with the exact query shape and parameters so phase 7 runs the identical one, timestamped before exposure. A tagged observation from the person may replace the query when its date, window and method are stated. A zero needs an observed zero or a documented first-ever measurement of a genuinely new metric; an existing product with missing history has an unavailable baseline, and the gate holds until it is measured or explicitly overridden. Capture the phase 4 guardrail baselines too. Where the spec declares model output, score the production sample against the eval rubric and add every failing output to the eval dataset, must_pass for safety or compliance. Done when every baseline has a value, window, query and timestamp, or a stated gap.
  7. Confirm the measurement exists (gate 6.3): a saved query or dashboard for the success metric, named and findable by someone else. Done when its name and location are written.
  8. Write release notes (gate 6.4) for the person who will use the thing: what they can now do, in their words, free of component names, ticket numbers and internal vocabulary, the phase 5 demo linked under **Demo:**. Then What changes for you, for people already using the product: what moved, behaves differently or was removed, from the spec's Impact on what exists. Then Help: the help that answers what users will now ask, in each language of PRODUCT.md → Markets, updated or "nothing to change" with why. Then plan ## Launch with go-to-market step 6 (gate 6.14). Done when the three parts and the launch are written.
  9. Schedule the outcome review (gate 6.5): an owner and a date from the phase 3 kill criteria, anchored to the verified exposure (the plan may carry a provisional date); where they disagree, the kill criteria win, noted. Change track: a change with no outcome to measure omits Baseline, Measurement and Outcome review and closes at phase 6 (steps 6, 7 and 9 skipped); one that sets a review runs them and goes to phase 7. Done when owner and date are written, or the change track's omission is stated. 9b. Promote what the initiative settled (gate 6.9). Entities, states, invariants and business rules from DESIGN.md → Domain model go into the project's DOMAIN.md, each marked with this initiative; a visual direction chosen here goes into TECH.md → Design system (kit, where the tokens live, composition rules). Where a promoted file is new (the first DOMAIN.md, a design system section), add its pointer to AGENTS.md per agents-md. Done when ## Promoted lists each promotion and its destination. 9c. Get ready for the first incident (gate 6.10). Under ## Production readiness: **Errors seen by:** how an error reaches the team before a user reports it (a tracker with an alert, a watched log query); **Backup restored:** a restore of the stored data tested once, with its date, or "not applicable" with why; with model output, **Agent traces:** where every turn and tool call is kept and who reads a sample how often (agent-design). Then fill what TECH.md → ## Operations is missing for the product's stage: Privacy always (notice, sub-processors with their data processing agreements, deletion on request, data region); AI disclosure when the spec declares model output (people are told they are dealing with an AI system; in the EU, Article 50 of the AI Act); after PMF SLO, Incident response (who is alerted, how users are told, the postmortem that becomes an anomaly initiative) Cost per active user with its budget alert and Load tested at ten times the spec's capacity, plus Model degradation with model output (must-pass cases rerun on sampled production turns, and the alert threshold). Privacy and AI disclosure name what a professional should confirm. Done when the readiness lines are filled and Operations has every line its stage requires.
  10. Prepare and report readiness. Write .builderos/initiatives/{initiative}/06-release.md per the release template, headings, bold labels and tables exact. A completed plan is ## RELEASE READY and stays at phase 6; a future rollout date is a plan. Done when the artifact exists and its readiness is reported.
  11. Execute within authorization and verify exposure. The agent prepares; a person authorizes crossing into user exposure, separately from approving the plan, after seeing the demo. Record **Authorized by:** in Exposure verification with who and how (gate 6.8). With deployment authorized, perform the rollout (on a new product, opening the door on the production that phase 5 set up, TECH.md → Environments), observe the exposed version in the intended environment and capture a resolving data or document source; otherwise report readiness and the remaining authorization or access. Record an actual timestamp, the environment, version and verification; installed-app behavior is observed in the installed app. A release rule that must hold every time (production deploys wait for approval) goes in TECH.md → Technical constraints with its deterministic check. A problem after exposure runs bug-diagnosis, the step 3 rollback first when users are affected. Done when exposure is observed and recorded, or the missing authorization is reported. 11b. Wait for the first real use (product track, gate 6.11): online is not in use. Record under ## First real use the first person outside the team who did the primary action, with its data or doc tag; until then the initiative stays at phase 6, and the person is told who to invite and what to watch. Done when that use is recorded. 11c. Wait for the first payment (product track, for sale, gate 6.12). Record under ## First payment the first payment by someone outside the team, with its data or doc tag (the payment record, the invoice); trials and pilots are use, not payment. Done when that payment is recorded.
  12. Gate the verified release. Run gate 6 and record it with the outcome date and the person who authorized (scripts/bos.mjs record 6 --review-due YYYY-MM-DD --accepted-by "who" and the required judgments; a change closing at phase 6 records without --review-due). Update project context and the roadmap. Done when gate 6 is recorded and the initiative is at phase 7 or closed.

Watch for

DoInstead of
Capture and timestamp the baseline before the first userA baseline captured after exposure
Measure missing history or mark it unavailableFilling it as zero
RELEASE READY until exposure is observedA rollout plan labeled SHIPPED
Ask for exposure authorization and record who gave it (6.8)Deploying because the plan was approved
Full release with a working rollback for 40 users, saying whyA percentage rollout on a tiny product

Capabilities

CapabilityUsed forFloor if absent
analytics.querySuccess-metric and guardrail baselinesA sourced observation with window/method; otherwise mark unavailable and name the measurement gap
analytics.eventsProduction events arriving before exposureThe production emission path in code, tagged as weaker evidence
db.queryBaselines in the application databaseSame floor: ask
repo.readFlag system, migration, deploy configurationRead TECH.md → Environments; the rollback is documented once you can describe it
logs.queryWhere production errors surface (step 9c)Ask how errors reach the team today; "nobody sees them" is the finding
docs.writePublishing release notes where users read themDeliver to the requested destination within existing authorization; report unresolved access
files.read / files.writeArtifacts, stateRequired

Gate 6.2 accepts a sourced manual measurement. Unknown history is a blocking gap; a measured baseline and actual exposure come from observation, and source-code inspection supplies neither.

Output Contract

## RELEASE READY for preparation; ## SHIPPED only for observed exposure backed by its evidence and a passing gate. A partial rollout is reported as partial.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.