Orchestrata Ultimate › Gates › wiring

Gate: wiring

At a glance

What it refuses

An artifact this branch added but left inert: a .claude/hooks/ file registered nowhere, a git-hook script the installer never references, a new src/ module reachable from no production entrypoint, a ruling-declared gate script no runner invokes.

Why it exists

Module docstring, quoting the ruling verbatim and the incident:

"Wiring gate — RUL-023 ('when I ask for a feature that means wiring the damn thing not just code vomit', verbatim, 2026-07-27).

See ~/workspace/knowledge-wiki/rulings/RUL-023-feature-means-wired.md for the ruling text and the incident (in a single session three capabilities were built and reported complete while inert: a git hook whose core.hooksPath pointed elsewhere, a .claude/hooks/ script registered only in a worktree's tracked settings.json, and a gate no runner invoked). RUL-012 already defines done as wired+deployed; this mechanically checks it for the artifact classes that are easiest to leave inert."

The four checks

How it runs

• Command: python scripts/check_wiring.py [--repo-root PATH] [--rulings-dir PATH] [--json]
• When it fires: the scripts/gates.py battery (registered as wiring); scripts/health_check.py.
• Scope: "everything this gate flags is scoped to the current branch's diff against master", so it grades what THIS change added, never the tree's accumulated history. No diff against master → exit 0, nothing to grade. Since 5740b332c (2026-08-12) the diff is no longer the commit range master...HEAD alone. diff_entries() unions three sources against the merge base of master and HEAD: (1) <merge-base>..HEAD, (2) git diff --name-status <merge-base> for staged and unstaged work, and (3) git ls-files --others --exclude-standard for new untracked files, graded A. Its docstring records why: with the commit range only, "147 files changed vs master, and the gate graded ZERO of them" (measured 2026-08-12).
• RULINGS_DIR (for W4) is resolved via the same mechanism as rulings_lint.py ("imported from there, not reimplemented") — "a machine without the knowledge-wiki clone must not go false-red: W4 is skipped, W1–W3 still run."
• Inputs: fingerprint roots _SRC = ("src", "scripts", "tests", "pyproject.toml") (the GATES entry "wiring": ([sys.executable, "scripts/check_wiring.py"], _SRC) in scripts/gates.py, read through the AST toolkit on 2026-10-23). An earlier version of this page said a 2026-08-28 audit dropped tests and added .claude/settings.json and .claude/hooks; the registry carries neither change.
• Caching — no git state, no .claude roots. An earlier version of this page described a {".py", ".sh", ".json"} mask plus a GATE_GIT_STATE HEAD/master/merge-base string mixed into the fingerprint. Neither exists: git log --all -S GATE_GIT_STATE -- scripts/gates.py and the same search for GATE_MASKS find no commit that ever contained them (checked 2026-10-23). _fingerprint() hashes only file paths and bytes with suffixes .py, .toml, .pl, .json, .md, .sh under the roots above. Consequences, derived from the code and not measured by a run: after a cached PASS, (1) a new .claude/hooks/*.sh or a .claude/settings.json edit does not re-run the gate, though W1 reads exactly those; (2) a new commit, a fetch or master moving does not re-run it either, though the merge base decides what it grades.
• Disclosed residual blind spot (registry comment): W4 also reads the EXTERNAL rulings directory — "a SEPARATE git repository from this one… No root in THIS repo's fingerprint can ever see a ruling added or edited there… Left undecided deliberately."

Typical failures and how to fix them

• Exit codes: "Exit 0 clean (or skipped, with a reason). Exit 1 on any violation, one line per finding: check id, artifact path, and the exact wiring site it is missing. Not a git repository, git unavailable, or no divergence from master: exit 0 with a printed reason, never crash."
• State over time: on 2026-08-29 the cached battery verdict was ok: false (a branch carrying unwired additions); the primary checkout's .cache/gates.json, last written 2026-09-22 10:01, records ok: true.
• Un-blocking: wire the thing — register the hook, reference the script in the installer, connect the module to an entrypoint, invoke the gate from a runner. The fix is never to delete the check.
• W1 and RUL-035 pull in different directions. check_w1_claude_hooks() still requires a new .claude/hooks/ file to be registered "under a PreToolUse/ PostToolUse/Stop matcher in .claude/settings.json". RUL-035 / ADR-0070 put hook registration at the user level (~/.claude or the orchestrata plugin) and never in a repo's .claude/settings.json. This run did not find a recorded resolution; check the ruling before adding a repo-level registration just to satisfy W1.
• A hook or settings change that the battery reports as unchanged, or a branch rebased onto a newer master. Because of the caching gaps above, run the gate fresh: python scripts/gates.py --only wiring --all, or the command directly.

Wiring (of the wiring gate itself)

• Invokers: scripts/gates.py (battery), scripts/health_check.py.
• Tests: tests/scripts/test_check_wiring.py.

Related

• Gates master index
• Development workflow
• RUL-023 — A feature request means the thing is WIRED
• RUL-012 — Done means wired and deployed
• RUL-035 — hooks at user level only and ADR-0070
• ADR-0042
• feature-loss — "code that exists but was never CONNECTED, not code that was DELETED after being connected" (that gate's own demarcation of this one)
• plugin-drift — wiring's delivery-path sibling: registered here, mirrored there
• Session-hooks family — the artifacts W1 protects