Wiring Test Skill Skill

Wiring Test Skill

Overview

Tests pass and code compiles, yet the feature is never wired up — command not registered, endpoint not mounted, module not imported, component never rendered. This is the exact failure loom's goal-backward verification exists to catch. Use this skill to write strong verification fields for a stage.

⚠️ truths is GONE. It was removed as a standalone field. Behavioral commands now live in acceptance; the goal-backward layers are artifacts, wiring, wiring_tests, dead_code_check. Any plan still using a top-level truths: block is stale — serde silently ignores it, so the "check" runs NOTHING and false-passes.

The five verification fields

| Field | Type | Proves | Timeout | | ----- | ---- | ------ | ------- | | acceptance | Vec<AcceptanceCriterion> | Build/test/lint AND observable behavior (the old truths commands) | 5 min (Simple) / 30 s (Extended) | | artifacts | Vec<String> (globs) | Files exist with real implementation | — | | wiring | Vec<WiringCheck> | Static connection point present (regex in a file) | — | | wiring_tests | Vec<WiringTest> | Runtime integration: command output matches criteria | — | | dead_code_check | Option<DeadCodeCheck> | No orphaned code (see /loom-dead-code-check) | — |

loom check <stage-id> [--suggest] runs the goal-backward layers (artifacts, wiring, wiring_tests, dead_code_check). acceptance runs during loom stage complete / loom stage verify.

Requirement (enforced by loom init and loom plan verify): every standard and integration-verify stage must define acceptance OR at least one goal-backward check. Knowledge stages are exempt.

CRITICAL PATH RULE: all paths (artifacts, wiring.source) are relative to the stage's working_dir. working_dir: "loom" + src/x.rs → .worktrees/<stage>/loom/src/x.rs. Never ../. Double-path (loom/loom/...) is the classic mistake.

YAML WARNING: never put triple backticks inside a description field — breaks the parser (often surfaces as a misleading "missing artifacts/wiring" error). Use plain indented text.

The one rule that matters: verify the CONSUMER, not the PRODUCER

A wiring pattern that greps where a symbol is declared / exported / imported passes while the feature is still unwired — the exact trap, committed inside the verification field. Grep the call / mount / render / dispatch site that proves the symbol is USED.

| ❌ Producer (exists ≠ wired) | ✅ Consumer (proves reachable) | | --------------------------- | ----------------------------- | | pattern: "mod new_command" | source: "src/cli.rs", pattern: "NewCommand =>" (dispatch arm) | | pattern: "pub fn handler" | source: "src/routes.rs", pattern: "/features.*create_feature" (route registration) | | pattern: "export function Foo" | source: "src/pages/Home.tsx", pattern: "<Foo" (render site) |

Pair every wiring entry with a behavioral acceptance command where one exists — observable behavior is the strongest wiring proof.

Field mechanics (how loom actually evaluates each)

wiring — `WiringCheck { source, pattern, description }`

pattern is a regex (Rust regex crate, RegexBuilder), passes if it matches anywhere in source. A missing/unreadable source file is a gap.
⚠ ! is a LITERAL character, not negation. You cannot express "must NOT contain" in wiring — use dead_code_check or an acceptance command with ! shell negation for that.
⚠ Match all visibility modifiers with pub.*fn name, not pub fn name — pub(crate) fn, pub(super) fn, and bare fn all differ.
Escape regex metacharacters you mean literally: Vec<String>, foo(), a.b — . ( ) < [ * + ? | \ are all special. Prefer anchoring on a distinctive substring over a fragile full-signature regex.

artifacts — `Vec<String>` glob patterns

A file is a gap if it is missing, empty after trimming whitespace, or contains a stub pattern:

TODO   FIXME   unimplemented!   todo!   panic!("not implemented
pass  # TODO   raise NotImplementedError   throw new Error("Not implemented

⚠ Markdown files (.md/.mdx/.markdown) skip stub detection — they legitimately contain "TODO" in prose.
⚠ Stub matching is a plain substring scan, so "TODO" inside a string literal or comment in a real source file trips it. If your implementation must contain the literal text TODO, don't list that file as an artifact.
There is no minimum byte size — the check is non-empty-after-trim, not ">100 bytes". Empty-file and stub gaps are distinct (ArtifactEmpty vs ArtifactStub).
Point at implementation files, not directories or bare test files.

wiring_tests — `WiringTest { name, command, success_criteria, description }`

The runtime, structured cousin of a behavioral check — a goal-backward layer (unlike acceptance). success_criteria (SuccessCriteria) fields, all optional:

wiring_tests:
  - name: "health endpoint responds"
    command: "curl -sf localhost:8080/health"
    success_criteria:
      exit_code: 0
      stdout_contains: ["ok"]
      stdout_not_contains: ["error"]
      stderr_contains: []
      stderr_empty: true
    description: "Health route mounted and reachable"

Use wiring_tests (not acceptance) when you want a runtime check counted as goal-backward proof and surfaced by loom check.

acceptance — Simple or Extended

acceptance:
  - "cargo test"                          # Simple: exit 0, 5-min timeout
  - command: "loom check st --suggest"    # Extended (TruthCheck): 30-s timeout
    stdout_contains: ["PASS"]
    stdout_not_contains: ["panic"]
    stderr_empty: false
    exit_code: 0

Templates by feature type

CLI command

acceptance:
  - "myapp new-cmd --help"                 # command responds
  - 'myapp new-cmd --help | rg -q "usage"' # primary use case wired
artifacts:
  - "src/commands/new_cmd.rs"
wiring:
  - source: "src/cli/dispatch.rs"
    pattern: "Commands::NewCmd"            # CONSUMER: dispatch arm, not `mod new_cmd`
    description: "Command dispatched in CLI"

API endpoint

acceptance:
  - 'curl -sf -X POST localhost:8080/api/features -d ''{"name":"t"}'' | rg -q id'
artifacts:
  - "src/handlers/features.rs"
wiring:
  - source: "src/routes/api.rs"
    pattern: 'post.*/features.*create_feature'   # route registration (consumer)
    description: "POST /features mounted"
wiring_tests:
  - name: "features endpoint reachable"
    command: "curl -sf -o /dev/null -w '%{http_code}' localhost:8080/api/features"
    success_criteria:
      stdout_contains: ["200"]

Module / library

acceptance:
  - "cargo test auth::"
artifacts:
  - "src/auth/mod.rs"
  - "src/auth/jwt.rs"
wiring:
  - source: "src/orchestrator/core.rs"
    pattern: "use crate::auth"             # a REAL consumer, not just `pub mod auth` in lib.rs
    description: "Auth used by orchestrator"

UI component

acceptance:
  - "bun test FeatureCard"
  - "bun run build"                        # build catches asset/CSS wiring a unit test misses
artifacts:
  - "src/components/FeatureCard.tsx"
wiring:
  - source: "src/pages/Dashboard.tsx"
    pattern: "<FeatureCard"                # rendered (consumer), not just exported from barrel
    description: "FeatureCard rendered in Dashboard"

Good vs bad

# ❌ proves nothing: tests pass with the feature unregistered; `src/` matches anything
acceptance: ["cargo test"]
artifacts:  ["src/"]
wiring:     []

# ✅ proves the feature is reachable end-to-end
acceptance:
  - "loom check st-1 --suggest"
  - 'loom check --help | rg -q "suggest"'
artifacts:
  - "src/commands/verify.rs"
wiring:
  - source: "src/cli/dispatch.rs"
    pattern: "Commands::Check"
    description: "Check command dispatched in CLI"

Realizability — a green check must actually PROVE something

A check that passes while asserting nothing is worse than none — it reads as "covered." Before finalizing, put each check through four gates:

Expressible — the harness can already do this. "Stub the response / seed this store" is not free; confirm the suite has the mechanism or add it as explicit work.
Executes the code under test — the runtime that runs the check loads the code asserted. A wiring grep only proves the call site exists in the file; it does not run it. Any change with real logic needs an acceptance/wiring_tests command that RUNS it.
Assertion strength matches the claim — a stdout_contains substring cannot guard a "byte-identical" contract; a presence check cannot guard behavior.
Actually selected — for EACH artifact a stage produces, at least one acceptance/wiring_tests command must FAIL if that artifact were broken. A test a CI filter never selects is dead coverage.

Final checklist

[ ] No stale top-level truths: block anywhere (removed field — silently ignored, false-passes)
[ ] Every wiring targets a CONSUMER site (call/mount/render/dispatch), not a declaration/export/import
[ ] wiring patterns are valid regex, metacharacters escaped, pub.*fn for visibility; no !-negation assumed
[ ] artifacts are specific implementation files that contain no stub text and aren't markdown-exempt where you needed stub detection
[ ] At least one acceptance or wiring_tests command exercises the primary use case end-to-end
[ ] Standard/IV stage has acceptance OR ≥1 goal-backward check (else loom init rejects it)
[ ] All paths relative to working_dir; no ../; no double-path
[ ] No triple backticks inside any description
[ ] For each artifact, some command would FAIL if it were broken

Agent Skills: Wiring Test Skill

Install this agent skill to your local

Skill Files