agentleFS
awebaiawebai/aweb/.claude/skills/cross-repo-change/SKILL.md

cross-repo-change

Skill129 starsChanged 6 days agoMITView on GitHub

Coordinate changes that touch both the aweb OSS repo and the hosted ac repo. OSS lands first; the hosted repository owns dependency derivation and deployment.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    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. Distinctive:

    remove an accepted shape,

    line 25
  2. Distinctive:

    Removal is a separate, later change: Remove support for an old shape only after MEASURING that nothing sends it.

    line 36
  3. Distinctive:

    Do NOT land the cloud change before the OSS release.

    line 85
  4. Distinctive:

    Do NOT use editable/sibling source installs as a permanent workaround.

    line 87
  5. Very specific:

    Do NOT accept both old and new formats "during transition" when both sides deploy atomically.

    line 89
  6. Distinctive:

    Do NOT change a contract without enumerating what currently sends it.

    line 97
  7. Distinctive:

    When reviewing a cross-repo bundle, grep ALL OSS migration trees touched in the bundle, not just the first one you find.

    line 120

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Cross-repo change coordination

Changes that affect both aweb (OSS) and ac (hosted SaaS) must be coordinated carefully because ac embeds aweb as a PyPI package.

Read this first: shippability

A released client is a permanent constraint. Once a client that sends X is in the field, the server must accept X until you can prove nothing sends it.

Everything below about atomic deploys applies to aweb ↔ ac only. Every other consumer — the channel plugin, pi, the aw CLI, self-hosted servers — updates on its own schedule. A running agent keeps its loaded module until the process restarts, which for a long-lived agent may be never.

Default: additive only. You may ADD an accepted shape. You may NOT:

  • remove an accepted shape,
  • make an optional field required,
  • reject a shape that is in the field,
  • narrow a type or add a stricter validator to an existing field.

The only exception: you may narrow a contract if you can NAME every consumer and show each deploys atomically with this change — same image, same deploy, same instant. "We will release the client right after" is not atomic. Put the justification in the commit message. Today exactly one boundary qualifies: aweb ↔ ac.

Removal is a separate, later change. Remove support for an old shape only after MEASURING that nothing sends it. If you cannot measure it, you cannot remove it.

Before merge, answer: what is deployed today, and does this still work against it? Check against PUBLISHED ARTIFACTS — npm pack the tarball and read the shipped bundle, or read the released tag — never against your own source. Your source and the published client disagree; that disagreement is the entire risk.

Why this section exists: on 2026-07-26 a chat mark-read change made a new field required and explicitly rejected the old one. Every published client sent the old field, so deploying it would have broken every client in the world. It was faithful to the anti-pattern below, which is correct for the cloud boundary and catastrophic for clients. The process was not ignored — it was followed.

Principle (aweb ↔ ac only)

The cloud imports aweb. Both deploy in the same Docker image. There is no transition period — when the cloud pins a new aweb version, both the old and new code deploy atomically.

This is the ONLY boundary with that property. Do not generalise it.

Sequence

  1. Design — agree on the contract change between OSS and cloud. Identify which side goes first.
  2. OSS first — land the OSS change on aweb main after the relevant focused tests pass.
  3. Release OSS — run make release-candidate TAGS='...' in aweb, then push each resulting tag separately. docs/release.md is authoritative. Wait for the exact package versions to be public.
  4. Cloud pins — update and review backend/pyproject.toml and backend/uv.lock to the exact public OSS versions on AC main.
  5. Cloud candidate — land the cloud-side source change on AC main, then run make release-candidate VERSION=X.Y.Z. Push the resulting vX.Y.Z tag; AC never publishes OSS artifacts.
  6. Verify — cloud tests pass against the real aweb package (not editable/sibling source).

Anti-patterns

  • Do NOT land the cloud change before the OSS release. The cloud CI will fail on import errors.
  • Do NOT use editable/sibling source installs as a permanent workaround. They mask version pinning issues.
  • Do NOT accept both old and new formats "during transition" when both sides deploy atomically. Pick one format. This applies to aweb ↔ ac and nowhere else. Before invoking it, confirm every consumer of the contract deploys in that same image. If any consumer is a published client — channel, pi, aw — or a self-hosted server, the rule INVERTS: accept both, prefer the new one, and remove the old only after measuring. See "Read this first: shippability".
  • Do NOT change a contract without enumerating what currently sends it. For the channel that check is small — its entire write surface is two calls, one of which has no body — so there is no excuse for skipping it.

Hosted schema mirrors

The ac repository maintains its own copy of the OSS schemas. The cloud's migration_paths.py points at backend/src/aweb_cloud/migrations/aweb/ and migrations/server/ — NOT at the OSS package's migrations in the installed .venv. So when an OSS release adds or alters a schema the cloud uses, the cloud needs a paired mirror migration with a matching ALTER/CREATE.

The migrations to mirror can live in DIFFERENT OSS tree subdirectories within the same release. Easy to miss the second one when walking commit by commit:

  • awid/src/awid_service/migrations/*.sql → mirror in cloud as migrations/aweb/.
  • server/src/aweb/migrations/aweb/*.sql → also mirror in cloud as migrations/aweb/ (same target dir, different OSS source dir).

When reviewing a cross-repo bundle, grep ALL OSS migration trees touched in the bundle, not just the first one you find. Pattern-blindness on the second mirror is the failure mode.

Each mirror file should be a one-statement ALTER/CREATE matching the OSS content, with a header comment naming the OSS source path so the lineage is explicit.

Example: aweb-aaje (proxy auth team_id format)

  1. OSS: changed X-Team-ID validation from UUID to colon-form, deleted _resolve_proxy_team_id (cde8889, 0fbe3d9, 78794ff)
  2. Released: aweb 1.10.1
  3. Hosted repo: pinned 1.10.1 and changed the bridge to send colon-form (2761d0a5, 1f6e4797)
  4. Both deploy together in the cloud Docker image

Notes

  • The hosted ac repo is cloned at ../ac relative to the aweb workspace.
  • Ownership and review routing live in the active team instructions and aw workspace status; do not rely on names written here.

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.