chore: adopt shared engineering guidance

.claude/settings.json

@@ -0,0 +1,6 @@
+{
+  "attribution": {
+    "commit": "",
+    "pr": ""
+  }
+}

.github/copilot-instructions.md

@@ -0,0 +1,19 @@
+# Copilot reviews
+
+For code review, follow `Code Review Rules` in root `AGENTS.md` and the local
+instructions that apply to changed files. Read the shared `origin89-review`
+skill before reviewing behavior, with its relevant domain skills.
+
+Use shared skills only when they are already available in the review context.
+The detailed criteria are maintained in
+[engineering's review skill](https://github.com/origin89hq/engineering/blob/main/skills/origin89-review/SKILL.md).
+If that skill is unavailable, follow the local review rules and disclose the
+missing shared context. Do not install tooling or assume a developer's ignored
+cache exists in the hosted review. A link is not evidence that a file was read.
+
+Review the requested head and its callers. Report a reproducible trigger,
+consequence, and precise code location. Do not invent findings, duplicate an
+existing finding without new evidence, or treat a CI pass as proof of safety.
+Leave formatting to the configured linters. Keep review comments concise, with
+one physical line per paragraph. A review request authorizes findings, not code
+changes or publication.

.github/pull_request_template.md

@@ -1,27 +1,11 @@
-## Problem and change
+<!-- Use a short conventional title, such as "fix: reject incomplete readings". Write each paragraph on one physical line; do not hard-wrap prose. Remove these hints when completing the description. -->
 
-Describe the problem and how the change fixes it. Include compatibility or
-migration details when relevant.
+## Change
 
-## Validation
-
-List commands actually run and their results. For changed nontrivial behavior,
-identify the distinct cases covered; add rows or explain why a category does not
-apply. Include further branches and faults where the contract requires them.
-
-| Behavior | Success | Invalid input | Boundary | Failure / recovery |
-| --- | --- | --- | --- | --- |
-| Describe the behavior | Test or evidence | Test or reason | Test or reason | Test or reason |
+<!-- Explain the problem and resulting behavior. Keep the PR focused on one result. Include a compatibility or rollout constraint only when relevant. -->
 
-Name any unperformed checks and remaining uncertainty. Include UI screenshots or
-hardware evidence when applicable.
-
-## When relevant
+## Validation
 
-Remove items that do not apply and complete the relevant ones:
+<!-- Name the commands actually run and their results. Summarize meaningful success, invalid-input, boundary, and failure cases in a sentence or short list. Use a table only when it makes the evidence easier to compare. State unperformed checks and material uncertainty. -->
 
-- New package, dependency, automation, or doc: why it is needed and how it stays current.
-- Dependency update: verified stable version, source, compatibility, and lockfile.
-- Package release: changeset and consumer impact, or why no release is needed.
-- Brand change: package version, exported assets/tokens, and visual checks.
-- Equipment behavior: safety-change record, fault evidence, and reviewer.
+<!-- When applicable, include UI screenshots, a Changeset or why none is needed, dependency compatibility, brand consumer impact, or the required equipment safety evidence. -->

.gitignore

@@ -5,3 +5,4 @@ __pycache__/
 /.origin89/engineering/
 /.agents/skills/origin89-*
 /.venv/
+/.claude/skills/origin89-*

.origin89/sync-engineering.py

@@ -141,33 +141,42 @@ def current_state(cache):
 
 
 def activate(project, cache, state, previous):
-    directory(project / ".agents")
-    discovery = project / ".agents" / "skills"
-    directory(discovery)
     old_names = set(previous["skills"]) if previous else set()
     new_names = set(state["skills"])
     targets = {}
-    for name in old_names | new_names:
-        link = discovery / name
-        target = os.path.relpath(cache / "current" / "skills" / name, discovery)
-        if link.is_symlink():
-            if os.readlink(link) != target:
-                raise ValueError(f"Refusing to replace a custom skill link: {link}")
-        elif link.exists():
-            raise ValueError(f"Refusing to replace local skill files: {link}")
-        targets[name] = target
+    # Check every discovery directory before changing links or the shared pointer.
+    for assistant in (".agents", ".claude"):
+        directory(project / assistant)
+        discovery = project / assistant / "skills"
+        directory(discovery)
+        for name in sorted(old_names | new_names):
+            link = discovery / name
+            target = os.path.relpath(cache / "current" / "skills" / name, discovery)
+            if link.is_symlink():
+                if os.readlink(link) != target:
+                    raise ValueError(f"Refusing to replace a custom skill link: {link}")
+            elif link.exists():
+                raise ValueError(f"Refusing to replace local skill files: {link}")
+            targets[link] = target
     next_pointer = cache / "next"
     if next_pointer.exists() or next_pointer.is_symlink():
         raise ValueError("Unexpected pending cache pointer; inspect it before retrying")
-    for name in sorted(new_names):
-        link = discovery / name
-        if not link.is_symlink():
-            link.symlink_to(targets[name], target_is_directory=True)
-    next_pointer.symlink_to(f"versions/{state['revision']}", target_is_directory=True)
-    os.replace(next_pointer, cache / "current")
-    for name in sorted(old_names - new_names):
-        link = discovery / name
-        if link.is_symlink():
+    created = []
+    try:
+        for link, target in targets.items():
+            if link.name in new_names and not link.is_symlink():
+                link.symlink_to(target, target_is_directory=True)
+                created.append(link)
+        next_pointer.symlink_to(f"versions/{state['revision']}", target_is_directory=True)
+        created.append(next_pointer)
+        os.replace(next_pointer, cache / "current")
+    except OSError:
+        # Before the pointer changes, remove only links created by this attempt.
+        for link in reversed(created):
+            link.unlink()
+        raise
+    for link in targets:
+        if link.name not in new_names and link.is_symlink():
             link.unlink()
 
 

AGENTS.md

@@ -1,10 +1,12 @@
 # Working in this repository
 
-At the start of each new task, run `just skills-sync` from the repository root.
+For hosted PR reviews, follow `Code Review Rules` below without running the local
+skills refresh. For other tasks, run `just skills-sync` from the repository root.
 Read `skills/origin89-working/SKILL.md` and the relevant domain skills under the
 immutable `path` printed by that command. Keep that snapshot for the task; do not
-refresh it halfway through work. Read local instructions and preserve stronger
-project constraints and project-specific skills.
+refresh it halfway through work. Before branch, commit, push, or PR operations,
+read `skills/origin89-commits/SKILL.md` from that snapshot. Read local instructions
+and preserve stronger project constraints and project-specific skills.
 
 If refresh reports cached content, continue with that verified cache and mention
 that the script could not check for updates. If no cache is available or
@@ -18,6 +20,12 @@ and exceptions below. Internal RFCs and research belong in
 [internal-research](https://github.com/origin89hq/internal-research). Add documentation
 only when its value and upkeep are clear; remove AI filler from every message.
 
+Confirmed problems left outside the current fix need an issue in the owning
+repository: search with `gh`, reuse a matching issue or create one with evidence,
+and return its URL. Follow the shared working skill's unfinished-work rule.
+Respect posting restrictions; if filing is blocked, provide the draft and say why.
+Finish authorized fixes instead of replacing them with backlog issues.
+
 ## Board work
 
 Read `CONTRIBUTING.md`, `board/CAMERA-LAYOUT.md`, and the board's Gerber
@@ -30,3 +38,21 @@ Run `just check` for the selected fabrication export. A failing copper rule is
 a board finding: report it and keep the gate failing until the artwork is fixed.
 Gerber validation does not establish electrical or bench qualification. This is
 a Python/CAD repo; do not add Node or Cargo workspaces without actual consumers.
+
+## Code Review Rules
+
+Read the shared `origin89-review` skill and relevant domain skills when available.
+In hosted review jobs that already provide `.origin89/engineering/skills/`, use
+that checkout without running the local refresh. If shared context is missing,
+review against the rules below and disclose that limit.
+
+- Preserve EasyEDA source and filed fabrication exports. Board changes need
+  the corresponding layout rule and checker evidence; keep copper failures visible.
+- Distinguish Gerber and placement checks from electrical or bench qualification.
+  Do not infer hardware safety from a passing host check or change artwork during review.
+- Trace findings through callers and guards. Give the trigger, consequence, and
+  precise location; distinguish completed checks from missing evidence. Leave
+  formatting to the configured tools and avoid duplicate or speculative findings.
+- Keep current PR defects in the review. Track confirmed pre-existing or explicitly
+  deferred problems as issues when filing is authorized; comments-only reviewers
+  provide a draft and state that it was not filed.

CLAUDE.md

@@ -1 +1 @@
-Read `AGENTS.md` first. Follow its task-start skill refresh and repository checks.
+@AGENTS.md

CONTRIBUTING.md

@@ -28,6 +28,10 @@ for working practices, tests, writing, and commits. `AGENTS.md` loads shared
 skills at the start of a task; `just skills-sync` refreshes them from engineering.
 Keep local constraints and domain-specific checks alongside those shared rules.
 
+Track confirmed problems left outside the current fix using the
+[shared issue rule](https://github.com/origin89hq/engineering/blob/main/skills/origin89-working/SKILL.md#track-unfinished-work).
+Use `gh` to find or create the issue, verify it, and return its URL.
+
 Install just 1.58.0 and Python 3.9+ for the skill bootstrap. Run `just --list`
 for repository commands and `just check` before opening a pull request.