authoring-ci-workflows

por posthog

Use when adding or editing a GitHub Actions workflow, composite action, or reusable workflow under `.github/` — new CI jobs, triggers, matrices, checkout/clone…

npx skills add https://github.com/posthog/posthog --skill authoring-ci-workflows

Descargar ZIP GitHub

40k

Authoring CI workflows

Before you propose a change to CI, check things already tried for the idea. It records what was measured, and why some good-sounding changes were reverted or rejected.

Conventions for .github/workflows/** and .github/actions/**. The linters own the mechanical rules (below); this skill is the judgment calls they can't enforce.

Before you write

Copy from a canonical file rather than from memory. ci-paths-filter.yml is the smallest complete example (triggers, concurrency, timeout, app token, Depot runner); ci-backend.yml is the reference for the heavy patterns (bounded-depth checkout, per-SHA concurrency, draft/ready, sharding).
Related skills — reach for these instead of duplicating them here:
- /gating-production-deploys — any job that pushes a prod image or dispatches a Charts deploy.
- /managing-github-actions-secrets — creating the GitHub App / secret a workflow reads.
- /depot-github-runners — Depot runner labels and sizing.
- /debugging-ci-failures — CI is red and you need to know why.

What the linters already enforce

Run bin/hogli lint:workflows and actionlint before pushing — they gate CI, and they (not this list) are the source of truth for what's enforced. Today that's: timeout-minutes on every job, the canonical PR concurrency block, a repo-wide budget for unscoped PR event dispatches, dorny/paths-filter negation safety, justification for full-depth checkouts, cache-write gating, semgrep service coverage, MCP path-filter coverage of the trees the MCP build compiles, required-check gate hygiene, secrets a reusable workflow reads being declared and passed by its callers, runner labels that name an OS version rather than a floating -latest alias, #-free values for the action inputs an inner shell re-parses, and generic GHA correctness (bad secrets.* / needs: refs, deprecated ::set-output, unknown runner labels). Third-party action digests are bumped by Renovate.

Check what a condition does before you push it

The linters check the shape of an if:; tools/workflow-plan checks what it does. It evaluates every job and step condition with GitHub's own expression evaluator against a synthetic draft PR, ready PR, fork PR, merge-queue run, master push, hourly schedule, and manual dispatch, and prints which jobs run:

hogli ci:plan .github/workflows/ci-backend.yml            # one row per job, one column per scenario
hogli ci:plan .github/workflows/ci-backend.yml --steps changes
hogli test:workflows                                       # the pinned expectations, run in ci-lint-workflows

tools/workflow-plan/tests/workflows.test.ts pins the rules in this file as runs / skipped rows per scenario: drafts skip the product matrix, the queue takes the full one, forks skip telemetry, a no-ci draft still reports its gate, the hourly run takes the matrices and skips the PR-only checks, and a superseded run records the gate as cancelled. Every job in those two workflows that carries an if: must be named by at least one row, which a coverage test enforces: add a conditional job and the suite fails until a row says what it should do. When you change a condition in ci-backend.yml or ci-frontend.yml, run the suite and update the row that describes the behavior you changed; when you add a lever to another heavy suite, add rows for it. The planner stubs the paths filter as "everything changed" and knows nothing a run: body produces, so a scenario stubs selector outputs by step id. It does not model trigger paths:, concurrency, or matrix expansion beyond a cell count; see the README.

The dispatch budget (500 runs / 10s / repo)

GitHub caps workflow-run dispatch at 500 runs per 10s per repo; overflow fails as startup_failure and takes unrelated runs in the same window down with it (a stack restack pushing many branches is the usual trigger). Minimize runs dispatched, not just work done — draft status doesn't help, runs dispatch before skip logic applies.

A reusable-workflow call counts as one run. Small always-fire PR workflows should be jobs under a single workflow_call parent, not their own dispatches (see pr-updated.yml / pr-opened.yml folded behind their parent — fold pr housekeeping into one dispatch). Event-type scoping moves to job-level if: guards:
```
jobs:
  turbo:
    if: contains(fromJSON('["opened", "synchronize", "reopened"]'), github.event.action)
    uses: ./.github/workflows/ci-turbo.yml
```
Prefer a trigger-level paths: filter over dispatch-then-skip: a run that only starts to no-op still spends a dispatch (gate container workflows on trigger paths).
```
on:
  pull_request:
    paths:
      - '.github/workflows/ci-x.yml'
      - 'path/to/product/**'
  workflow_dispatch:
```
Judgment call — trigger paths: vs a runtime dorny/paths-filter job. Use trigger paths: for a workflow that is skippable as a whole. Never put a trigger paths: on a workflow whose check is required by branch protection: a required check that doesn't dispatch on a PR leaves it stuck "waiting for status" and unmergeable. Keep those firing on every PR and gate internally with a changes job (also the right call when several jobs branch on different path sets). Heavy matrices (ci-backend, ci-nodejs) do exactly this — deliberate.
Delete dead dispatchers outright. A disabled-but-still-triggered workflow keeps dispatching no-op runs against the cap — remove the trigger, don't just disable it.
A filter has to cover what the job actually compiles, not just where the code lives. A build that reaches out of its own tree — through a tsconfig paths alias, a workspace dependency, a generated file — breaks when a filter lists only the service directory. The PR touching the imported tree skips the job, the merge-queue run skips it too, and the break lands on master, where it then fails every later PR whose diff does match. Cover the enclosing directory, not the imported file: the module drags its own relative imports along. When the imported tree is busy, give it a second filter output and gate only the jobs it can actually break, rather than widening the one that also fires an expensive suite. WF009 derives the MCP filters' required trees from the MCP sources and fails when one is uncovered; a build with the same shape wants the same treatment.

Concurrency

Every PR-triggered workflow gets the canonical block:

concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref || github.ref }}
  cancel-in-progress: ${{ github.event_name == 'pull_request' }}

Cancel superseded PR runs; never cancel across master pushes. WF002 rejects a bare cancel-in-progress: true on any push-triggered workflow. Where latest-wins is genuinely right (a cache warmer), say so with # hogli-lint: allow-master-cancel -- <reason>.
Use github.ref as the fallback, never github.run_id — run_id is unique per run, so it silently gives every push its own group and dedup is lost.
Publish-on-push workflows must not let two master pushes race :latest / a deploy dispatch. Key the push arm per-SHA (see ci-backend.yml):
```
group: ${{ github.workflow }}-${{ github.event_name == 'push' && github.sha || github.head_ref || github.ref }}
```

Required-check gates

The "gate" is the collate job that emits the required status check by reading needs.*.result. By convention its display name ends in Pass (Django Tests Pass, Visual regression tests pass), but WF007 also finds gates structurally when a step reads needs.<dep>.result, because the convention is not universally followed. A job that inspects results without gating anything opts out with # hogli-lint: not-a-required-gate — <reason> above the job key. Gates and the workers they inspect share the same condition:

Job	Condition	Why
Gate	`if: ${{ !cancelled() }}`	It emits an explicit verdict on every completed run, and a superseded run records `cancelled`, not `failure`.
Workers	`if: !cancelled()`	So a superseded run actually stops instead of holding the concurrency slot.

The gate condition must contain !cancelled(), with optional ${{ }} wrapping. always() is rejected: it is identical to !cancelled() on any run that is not cancelled, but on a superseded run it runs the gate after the cancel and reports failure, which inflates every CI failure-rate metric with runs a developer merely pushed over.

Extra predicates may only be OR-ed on, never AND-ed. This is a correctness rule, not a style one. A conjunction gives the gate a second way to be false, and a job skipped by its own condition records skipped, which branch protection reads as a pass. Both conclusions occur in the same cancelled run: in run 33496887370 Calculate running time recorded cancelled while Backend coverage report recorded skipped, because an AND-ed predicate of its own was already false. A disjunction cannot be false while !cancelled() is true, so cancellation stays the gate's one false predicate and the conclusion stays cancelled.

Cancellation still fails closed. A gate on !cancelled() that never starts records conclusion cancelled, never skipped. Measured on a superseded run (evidence): the gate recorded cancelled with zero steps, while the always() control ran after the cancel and recorded failure. GitHub's status checks reference lists success/neutral/skipped as passing and never places cancelled among them, and a commit whose only checks are cancelled rolls up to FAILURE (evidence). That is inference rather than a documented guarantee, which is the reason for the next rule.

A workflow that cancels its own run must OR that signal onto its gate. ci-backend cancels itself when repo checks or OpenAPI types fail deterministically, to stop paying for runners on a failure a retry cannot fix. Under a bare !cancelled() those real failures would report cancelled too, which both hides them from the failure-rate metric and rests merge safety on the inference above. OR-ing the deterministic-failure output back on keeps the honest verdict, because the disjunct is true, so the gate dispatches despite the cancel:

if: >
  !cancelled()
  || needs.repo-checks.outputs.deterministic_failure == 'true'
  || needs.check-openapi-types.outputs.deterministic_failure == 'true'

Measured on a self-cancelled run (evidence): the bare !cancelled() gate recorded cancelled, the OR-ed gate ran and recorded failure. Only superseded runs then report cancelled, and every real failure keeps a failure conclusion.

A failure-rate metric keyed on a gate job must exclude cancelled. Only success and a decisive failure are a verdict, so a denominator that counts cancelled measures push behavior, not test health. Find those rows through the run's conclusion, not the gate job's. The gate job's own conclusion changed on 2026-09-04: a superseded gate recorded failure before that date and records cancelled after it. A metric that drops the superseded rows from the numerator and the denominator stays comparable across that date. One that filters on the gate job's conclusion alone does not. The run's conclusion lives in the warehouse table github_workflow_runs. The posthog-ci-running-time event cannot supply it: the action fills that event's conclusion property from the job named in its status-job input, and every caller passes a gate job name. It writes the same value to the workflow_run group, so both of those fields carry a gate conclusion under a run-shaped name. A metric keyed on jobs joins github_workflow_jobs to that table on run_id, then scopes the job side to a single run_attempt. The runs snapshot keeps one row per run id, at its newest attempt, so an unscoped read stamps that conclusion onto every earlier attempt's gate and counts the gate once per attempt. Do not enforce that scope by joining run_attempt equality: it blanks or drops every earlier attempt, which is the population that actually ran after a partial re-run (products/engineering_analytics/backend/logic/views/job_costs.py records that decision). Reuse the canonical predicates instead of writing a new denominator: CONCLUSIVE_RUN_CONDITION in products/engineering_analytics/backend/logic/queries/_workflow_filters.py, and computeHealthSummary in products/engineering_analytics/frontend/lib/runHealth.ts. That run-level key identifies superseded runs only where the workflow never cancels its own run. Where it does (the rule above), a deterministic failure records run conclusion cancelled too, so the canonical predicates drop that honest failure together with the superseded rows. Measured on Backend CI run 34204389260: the run recorded cancelled while the Django Tests Pass gate recorded failure. Keep those rows in the numerator and the denominator, and find them through the cancel jobs: each one dispatches only on its deterministic-failure signal, so a success from any of them marks that population on both sides of 2026-09-04.

Four rules for the gate body:

Allowlist every dependency, never denylist. Assert success/skipped and fail everything else. A dependency tested only against == 'failure' lets cancelled through, and one bad dependency is enough — a gate that clears four correctly and one with a bare failure test is still wrong. The trap is the changes detector: clearing it with == 'failure' and then reading needs.changes.outputs.* reports green on cancellation, because those outputs are empty and the gate takes its "nothing to test" exit.
needs every job that produces coverage, and every upstream that can skip one. If a job's failure would only cascade into a downstream job being skipped, the gate reads that as a pass and you get a green check with zero tests run. The usual shape is a changes detector one step above the suite: it fails, the suite skips, and the gate reports success having run nothing (measured on ci-nodejs). Name every job whose failure would skip one you do test, not only the direct ones. Stop there. A test selector whose failure leaves the suite running in full is not gate-critical, and demanding it turns a recovered run into a red required check.
Legitimate skips must still pass. A frontend-only PR skips backend jobs by design.
Every dependency's result must reach a fail-closed allowlist guard. One inline if per dependency is the clearest form, but a shared shell helper or an env: block is equally fine: WF007 traces each result through assignments, ${!var} indirection, and helper argument positions within that step. The guard must compare with !=, join multiple allowed values with &&, and unconditionally exit 1 when entered. Comparisons in another step, comments, logs, or branches that do not exit nonzero prove nothing and are rejected. A result whose guard WF007 cannot follow is reported rather than assumed safe, so an unusual routing may need the checks moved inline.

WF007 enforces 1, 2, 4, and the !cancelled() condition, and it takes the dependency list from needs: as well as the step body, so a job you wired into needs: and then forgot to test is reported rather than silently trusted. For rule 2 it walks the needs: graph above each dependency and reports any job the gate does not test, which is the half a linter can see. It follows an edge only when the upstream's failure would actually skip the job below it. A job whose if calls no status function is skipped by any failed upstream, and so is one held behind success() or cancelled(), since neither is true in that state. A job that reaches always(), !cancelled() or failure() keeps running, and is skipped only where its own condition compares against the failed job and goes false: an output reads back empty, while result reads back failure, so a recovery path testing result == 'failure' still runs. The half it cannot see is a coverage job with no needs: edge into the gate at all: "reporting job" and "coverage job" look identical from outside the graph, so that one is on you and the reviewer.

What GitHub does with each conclusion

Rule 3 works because a skipped check run satisfies a required context, in the Trunk queue as well as on GitHub. Build Docker image rides on that: it concludes skipped on most PRs and they merge anyway.

Three cases behave in ways the name does not suggest:

A required context that no check run reports stays pending and blocks. A trigger-level paths: filter that silences the whole workflow produces exactly this.
A job-level continue-on-error: true posts a failure check run even though dependents read success and the run goes green. Use step-level continue-on-error plus an explicit verdict step instead.
A matrix that expands to zero cells fails its dependents on GitHub Actions and posts no check run at all. Guard any fromJSON matrix with an if: that skips the job when the list is empty.

Required contexts are pinned to one app: every entry in this repo's master ruleset carries integration_id: 15368, the github-actions app, so a check run from any other app never satisfies one however exactly the name matches. /depot-ci has the measurements, the rulesets query, and how Depot CI differs.

Checkout / clone — sparse first, then shallow

This repo is 45k tracked files and 4.6 GiB of packed objects, so what you materialize costs more than how much history you fetch. Measured checkout-step durations, from the GitHub API on real runs:

Pattern	depot-ubuntu-24.04	GitHub-hosted ubuntu
`sparse-checkout` of a few paths, cone mode off	0–7s	0–7s
plain checkout (depth 1)	11–13s	22–44s
`fetch-depth: 1000` + `filter: blob:none`	53–59s	—

Biggest lever: check out only the paths the job reads. Sparse-checkout is not just for single files — a job that runs a local composite action, reads a JSON config, or lints one directory should name those paths and nothing else.
```
- uses: actions/checkout@<sha> # v6
  with:
    sparse-checkout: |
      .github/actions/paths-filter
      .github/clickhouse-versions.json
    sparse-checkout-cone-mode: false
```
Always set sparse-checkout-cone-mode: false. Cone mode additionally materializes every file in the repo root — here 70 files and 21.5 MB, .test_durations alone 18.5 MB — which is most of what you were trying to avoid. Cone mode also only takes whole directories, so it drags in all of bin/ when you wanted one script.
filter: blob:none is counterproductive if the job then materializes the tree. It removes blobs from the fetch, but git checkout immediately lazy-fetches every blob in HEAD in a second round trip, which is slower than having fetched them in the pack. That lazy fetch also intermittently fails its per-blob credential lookup with could not read Username for github.com (#59779, blocked a merge until retried). Pair blob:none with sparse-checkout so the lazy fetch is a handful of blobs, or drop it and take the plain depth-1 checkout.
Default: plain actions/checkout (depth 1). Add nothing.
Diffing against the PR base: you need real history, so bound the depth, filter blobs, and sparse-checkout the files the job reads:
```
- uses: actions/checkout@<sha> # v6
  with:
    fetch-depth: 1000
    filter: blob:none
    sparse-checkout: .github/actions/paths-filter
    sparse-checkout-cone-mode: false
- name: Fetch PR base for affected diff
  if: github.event_name == 'pull_request'
  env:
    BASE_REF: ${{ github.event.pull_request.base.ref }}
  run: git fetch --no-tags --depth=1000 --filter=blob:none origin "$BASE_REF:refs/remotes/origin/$BASE_REF"
```
A sparse working tree does not affect git merge-base, git diff <a>...<b>, git log --name-status, git ls-tree, git ls-files, or git show <rev>:<path> — those read the object database or the index. Only commands that compare against the worktree (git diff HEAD, git status) see the skip-worktree entries. One caveat when blob:none is also set: git show <rev>:<path> still needs that blob, and a sparse checkout never downloaded it, so the read becomes a lazy fetch that can fail. Name any file a step reads that way in the sparse set — ci-dagster.yml does this for docker-compose.base.yml, whose contents feed a cache key.
changes / paths-filter gating jobs: on pull_request the vendored .github/actions/paths-filter diffs via the GitHub API and never touches the tree. The only reason to check out is that a local action must exist on disk, so sparse-checkout .github/actions/paths-filter plus any file the job's own steps read. Never pass base: HEAD to paths-filter from a sparse job — that routes it to git diff HEAD, which a sparse worktree makes return nothing, so every downstream job silently skips green.
Foot-gun: git fetch --deepen=N with no refspec falls back to the wildcard refs/heads/* and pulls every branch. Always pass an explicit, --no-tags, --filter=blob:none refspec scoped to the base ref. (Bumping actions/checkout's own fetch-depth is safe — it uses a scoped refs/pull/N/merge refspec.)
The linter rejects fetch-depth: 0 unless you add filter: blob:none, use sparse-checkout, or justify it with # hogli-lint: allow-full-depth-checkout -- <reason>. Genuinely full-history jobs: repo mirroring (foss-sync.yml), tag/submodule version math (release-cli.yml, desktop-tag.yml). Most base-diff jobs should use bounded 1000 + blob:none plus a sparse set.

Pinning and tool versions

Pin every third-party action to a full 40-char commit SHA with a # vX.Y.Z comment. A moved tag can ship malicious code; pinning is also reproducible and skips a per-run GitHub-API version lookup. The only sanctioned exception is a debug-only action. In-repo composites use a local path with no ref (uses: ./.github/actions/pnpm-install).
Node version comes from .nvmrc — node-version-file: .nvmrc, never a hardcoded node-version:. Sparse-checkout .nvmrc if the job has no checkout.
Pin setup-uv's version: — an unpinned setup-uv calls the GitHub API on every job and burns the rate limit.

Never splice a caller's input into a command an inner shell re-parses

A composite action that builds sh -c "... $INPUT" hands the container's shell a script, not a flag list. That inner shell re-parses it, and exec replaces the shell on the first command — so a token that ends a command there truncates every flag after it. With #, a newline or ; the truncation is silent: exec never returns, nothing else runs, and the job exits 0. & and | truncate the command too, but the leftover flag then runs as a command of its own and the step fails with 127.

The input reaches it looking innocent. Inside a YAML block scalar (args: >-) a # is data, not a YAML comment, and a more-indented line is not folded, so it keeps its newlines. Neither needs unusual input — a note or an indented flag is enough.

Measured on #101671 before this was fixed: semgrep-go loaded 5 of the 7 configs it listed, semgrep-rust 5 of 7, semgrep-general 4 of 6, and the --exclude-rule / --include scopes below the comment went with them. CI was green throughout.

Quoting harder is not the fix — the value already sits inside double quotes, and another pair collapses every flag into one argument. Split it yourself and pass positional parameters, the way .github/actions/semgrep-ci now does:

set -f                          # split on whitespace WITHOUT expanding globs
semgrep_args=($SEMGREP_ARGS)
set +f
... sh -c 'exec tool ... "$@"' sh "${semgrep_args[@]}"

Nothing re-parses the value: each word reaches the tool verbatim, a glob arrives unexpanded for the tool to match itself, and a stray # becomes an argument the tool rejects loudly instead of a comment that eats the rest.

WF012 watches for the pattern coming back. It flags any action that splices an input into an inner shell, and checks that action's callers until it stops. An empty SHELL_SPLIT_INPUTS is the healthy state. # stays fine in every other input — dorny/paths-filter filters:, actions/github-script script:, a webhook payload: — so the rule is per-input, never blanket.

Network fetches

Downloads from outside the runner need retries, or a transient reset becomes a red check with no findings (actionlint died on curl: (35)).

curl -fsSL --retry 5 --retry-all-errors --retry-max-time 60 --connect-timeout 10 -o "$out" "$url"

--retry-all-errors is the part that catches a reset; plain --retry covers only timeouts and 408/429/5xx, and --retry-connrefused adds ECONNREFUSED, not ECONNRESET.
Drop it on GitHub API calls: with -f it also retries 403 and 404, spending five more requests on an already-empty token bucket.
No --retry-delay (it replaces exponential backoff with a fixed wait). Keep -f, or an error page lands in your output file at exit 0.
Don't retry anything non-idempotent (webhook posts, telemetry), or where a shell loop or readiness wait already retries.

Tokens — dedicated App tokens for high-volume calls

GITHUB_TOKEN shares one ~15k req/hr bucket across every job of every run in the repo; it goes hot at merge peaks and change-detection jobs fail before real work starts. A dedicated GitHub App installation is its own bucket — rate-limit headroom plus blast-radius isolation.

- uses: actions/create-github-app-token@<sha> # v3.1.1
  id: app-token
  # forks can't read org secrets — fall back to github.token
  if: github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == github.repository
  with:
    client-id: ${{ vars.GH_APP_POSTHOG_PATHS_FILTER_APP_ID }}
    private-key: ${{ secrets.GH_APP_POSTHOG_PATHS_FILTER_PRIVATE_KEY }}

# a later step consumes the token (falling back to github.token on forks):
- uses: some-action@<sha>
  with:
    token: ${{ steps.app-token.outputs.token || github.token }}

Right-size, don't over-isolate. One heavy consumer (change detection on a hot matrix) deserves its own app; a long tail of light workflows can share GITHUB_TOKEN. Convention: GH_APP_<PURPOSE>_APP_ID (an org variable — app IDs are not sensitive, and org secret slots are capped at 100) + GH_APP_<PURPOSE>_PRIVATE_KEY (an org secret).
Cross-repo tokens set explicit owner: + repositories: (least privilege).
Creating the app + secret is out of scope here — use /managing-github-actions-secrets.

Secrets in reusable workflows

A workflow_call workflow receives no secrets on its own. A secrets.X it reads interpolates to an empty string unless it declares X under on.workflow_call.secrets and every caller passes it, or a caller uses secrets: inherit. Nothing fails when that happens: an App-token step under continue-on-error falls back to github.token and the check stays green, which is how ci-turbo silently lost its dedicated rate-limit bucket. WF010 fails an undeclared read, and fails a caller that omits a secret declared required: true. Declare required: false only when the callee genuinely works without the value, the way a smoke-test build withholds a publish key on purpose; that is the callee's promise, and the linter takes it at its word.

Forks and untrusted PRs (public repo)

Fork pull_request runs (and Dependabot) get a read-only GITHUB_TOKEN and no secrets. Make those runs pass, and never let untrusted code reach a secret.

Guard secret-needing steps with if: github.event.pull_request.head.repo.full_name == github.repository, and degrade rather than fail (|| github.token, or the raw test outcome).
Secret-injecting builds (BuildKit --secret, registry login) must skip forks — gate both the changes job and any always() build job (block fork PRs from rust image build).
Comment or label only on same-repo PRs — the fork token can't write.
To act on a fork PR with secrets/write (reviewer or label bots), use pull_request_target: base-repo permissions, but it must never check out and run fork code. That's why those workflows can't fold into a pull_request parent.
First-time contributors need maintainer approval before workflows run (action_required) — expected.
Depot CI runs no fork PR at all, so the backend router keeps fork PRs on GitHub Actions and the required check stays on the head. Never re-push a fork's head in-repo to get it a Depot run. See Pull requests from forks.

Timeouts

Every job sets timeout-minutes, sized ~2-3x observed max; gate/aggregation jobs get ~5m. The default is 6 hours — a hung job burns paid minutes silently. Caveat: timeout-minutes is invalid on a job that only uses: a reusable workflow — put the timeout inside the called workflow instead.

Caching

Route through the shared composites rather than hand-rolling actions/cache: ./.github/actions/pnpm-install (single pnpm-<os>-<lockhash> key, restore only; pnpm-store-cache.yml writes it on master), astral-sh/setup-uv with enable-cache: true, Depot cache via ./.github/actions/build-n-cache-image. One canonical key per artifact; gate saves to master or key deliberately per-ref.

pnpm on a depot-* runner uses Depot Cache, not GitHub Actions cache. Depot transparently handles the actions/cache API calls made by pnpm-install. pnpm-store-cache.yml warms two separate backends: Depot Cache on its depot-* matrix leg and GitHub Actions cache on its ubuntu-* leg. Never use GitHub's cache API, usage total, or 10 GB limit to diagnose pnpm caching on Depot jobs, and do not propose migrating those jobs to Depot Cache: they already use it. Inspect Depot's cache data for Depot jobs and GitHub's cache data only for GitHub-hosted jobs. Keep the pnpm writer master-only because Depot Cache is repository-scoped and has no branch isolation. PR-scoped writes on GitHub-hosted runners still fragment GitHub's 10 GB LRU cache.

Any job that runs manage.py migrate against a fresh Postgres must restore the master schema dump first, keeping the migrate as a seconds-long top-up. A from-scratch replay of the full migration history grows with every migration merged and already costs more than most jobs' timeout-minutes, so an uncached migrate is a timeout that hasn't fired yet (agent-skills cancelled at 30 min with the checks green). Copy the three steps (compute keys, actions/cache/restore, prime) from ci-agent-skills.yml for compose-stack jobs or ci-rust-flags-integration.yml for service-container jobs; hogli db:restore-schema-fresh reads TARGET_DB to pick the database. A miss falls through to the full migrate, so the restore is never a correctness risk. The only sanctioned exception is a job whose purpose is validating the migration history itself (ci-backend's check-migrations), where a restored dump would mask what it checks.

Runners

depot-ubuntu-<version>[-<vCPU>] for build/compute-heavy jobs (the -4/-8 suffix bumps CPU from the 2-vCPU default); GitHub-hosted for light jobs. Name the OS version, never ubuntu-latest. GitHub moves the -latest aliases to a new image on its own schedule (24.04 became latest in 2025, 26.04 follows), which changes the toolchain and system packages under every job at once with nothing in this repo to bisect. WF011 rejects ubuntu-latest, macos-latest, windows-latest and the depot-*-latest mirrors in runs-on and in strategy.matrix, so use ubuntu-24.04 (or depot-ubuntu-24.04) and land an image bump as its own PR. A runner matrix that an expression builds (fromJSON(needs.plan.outputs.val)) is opaque to the linter, so it fails closed: pin the labels where they are generated (dist-workspace.toml for the release workflows) and put # hogli-lint: allow-generated-runner-matrix -- <where they are pinned> above the job key. New Depot labels must be added to the allow-list in .github/actionlint.yaml or actionlint fails. Details: /depot-github-runners.

Draft vs ready-for-review

The merge gate is the merge queue's run on a trunk-merge/ branch. So a heavy suite runs its narrowed subset on drafts and on ready PRs, and only the queue run takes the full matrix. ci-backend.yml, ci-frontend.yml, ci-storybook.yml, and ci-e2e-playwright.yml all work this way. Exclude the queue with !startsWith(github.head_ref, 'trunk-merge/'), or with draft != true since those PRs open as drafts — not with draft == true.

Still add ready_for_review to the pull_request types — a no-ci draft skips the workflow outright, and that event is what gives it a run — and make aggregator "... Tests Pass" jobs treat skipped as success so drafts still report.

What draft state does decide is the fallback when a selection cannot be trusted (config change, oversized diff, selector crash):

draft → skip the suite. Its own ready run selects again, so nothing is lost.
ready → full matrix. No later run on that PR would cover it, and a narrowed run the selector can't vouch for is false confidence.

The real test is not draft-vs-ready, it is whether a later run on this PR exists to defer to. So draft → skip is only valid when the workflow lists ready_for_review in its pull_request types and its "... Tests Pass" aggregator treats skipped as success. Otherwise the fallback is full on drafts too: ci-nodejs.yml has a bare pull_request: trigger, so a skipped draft would never get a second run, and ci-e2e-playwright.yml skips drafts entirely, so there is no draft run to fall back from.

turbo-discover.js (draft ? 'skip' : 'full') and ci-frontend.yml's fall_back are the two reference implementations of the draft/ready split; ci-nodejs.yml and ci-e2e-playwright.yml are the reference for always-full. Foot-gun: if the job that selects tests is cancelled mid-flight, its mode output is empty — normalize empty-mode on a draft to skip, or the draft grabs the full matrix and serializes the ready run behind it.

Forcing the full matrix on a draft

The run-ci-backend and run-ci-frontend labels force the full matrix, but a label alone starts nothing. It takes effect on the next push, or when the PR is marked ready for review. An empty commit is enough:

git commit --allow-empty -m "chore(ci): run the full matrix" && git push

Do not add labeled/unlabeled back to a merge gate's on.pull_request.types to avoid that push. GitHub cannot filter a label trigger by name, so every unrelated label re-runs the full matrices against a commit CI has already covered. Guarding it inside the workflow is worse: skipping the gate job cascades to the if: !cancelled() aggregator, which counts a skipped dependency as success and posts a green required check with no tests behind it.

A selector needs telemetry, or nobody knows whether it bites

A narrowing that falls back on most runs looks identical in the YAML to one that works. The Playwright selector was eligible on 2,783 runs over two weeks and narrowed on 167 of them; the rest fell back, most often on one glob. That is unreadable without an event, so every selector emits one to the DevEx project (347861): posthog-ci-test-selection, posthog-ci-e2e-spec-selection, posthog-ci-jest-selection, posthog-ci-nodejs-selection.

Copy the shape from capture-jest-selection in ci-frontend.yml:

Emit on full runs too, tagged so they are distinguishable. They are the baseline a narrowed run is measured against.
Give the fallback a closed category plus an unbounded detail. The category is what you group by; the detail is what tells you which glob to attack.
Emit the counts from every branch of the selector, including the ones that narrow to nothing.
continue-on-error, github.run_attempt == '1', and the same-repo if: — telemetry never reds CI, never double-counts a re-run, and forks have no secret.
Do not duplicate timings. posthog-ci-running-time-job already carries each job's duration; join it on run_id.

What this still does not answer is whether a narrowed run would have caught what broke. Only backend scores that, in tools/test_selection_verdict.py, which reads the run's JUnit and reports recall. The cheap oracle for the rest is the queue: the trunk-merge/** run tests the same code with the full suite, so a failure there whose file was reachable from the diff and was not selected is a miss, at no extra compute.

The hourly master lane

The merge queue's trunk-merge/** run tests every commit before it lands, so re-running a heavy suite on the master push tests a commit that CI already covered. Those suites skip push and take their master coverage — and their Trunk flaky-test baseline — from an hourly schedule: instead.

Crons are offset so the runs do not all fire at once, and the offsets live here rather than in the workflows:

Workflow	Minute
`ci-frontend.yml`	7
`ci-nodejs.yml`	13
`ci-backend.yml`	23
`ci-dagster.yml`	33
`ci-python.yml`	43
`ci-mcp.yml`	53
`ci-backend-update-test-timing.yml`	17

Adding a seventh: pick an unused minute, add the row, and keep the gap at ten minutes.

ci-backend-update-test-timing.yml sits in the table too. It is one small job that merges the artifacts of the hourly runs, not a suite, so it does not need the ten-minute gap.

Give the cron its own concurrency group. cancel-in-progress is false outside pull requests, but GitHub still keeps at most one pending run per group, so a newer run replaces an older pending one. An hourly run that shares the ref group with master pushes therefore holds the group for its whole duration while pushes queue behind it, and each new push discards the previous pending one along with whatever per-commit checks it carried.

group: ${{ github.workflow }}-${{ github.event_name == 'schedule' && 'scheduled' || github.head_ref || github.ref }}

That keeps hourly runs queueing behind each other rather than stacking, and leaves push behavior untouched. ci-backend.yml reaches the same place from the other side: its push arm is already keyed per SHA, so pushes never share a group with the cron. Prefer the 'scheduled' key when the push lane still runs real per-commit work, because a per-SHA push arm also gives up the deduplication that collapses a burst of master pushes into one run.

The paths filter must be skipped on schedule, and every output it feeds must default to true. On a cron the action has nothing to diff against: it gets no base input, so base resolves to the default branch and equals head, and before is set only on push events. It falls back to the last commit alone, so one docs-only commit narrows the hourly run to nothing and reports green having tested almost nothing — a silent failure, not a red one. Guard the step with if: github.event_name != 'push' && github.event_name != 'schedule', and give each consumed output a || 'true' default.

Any step that reads a filter output needs the same schedule arm. ci-dagster.yml's build-matrix is the cautionary case: without it the matrix is [] and the hourly run passes having run no tests.

A lane that stops producing runs is invisible to the master-red alerter: it reads run completions, so a dropped cron reads as unreadable and drops out of evaluation rather than paging. Add every converted workflow to SCHEDULED_GATING_WORKFLOWS in ci-alerts-devex.yml in the same change, or its failures stop paging altogether.

Backwards-compat with unrebased PRs

A workflow edit hits every open PR the instant it merges (it runs against PR-merged-with-master), but companion changes — a new dependency, file, or config — only reach a branch when it rebases. If the workflow starts requiring something unrebased branches lack, every in-flight PR fails before its tests run. Make new behavior degrade gracefully when the prerequisite is absent, or gate it. Roll out a new blocking lint the same way: ship continue-on-error, clear the inbox, promote to blocking.