dead-code-audit
Systematic codebase audit for unwired code — identify unregistered routers, uninvoked hooks, orphaned components, and file discovery issues. All findings become "wire it in" issues — never deletion issues.
mrveiss · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add mrveiss/AutoBot-AI --skill dead-code-audit技能说明
name: dead-code-audit description: Systematic codebase audit for unwired code — identify unregistered routers, uninvoked hooks, orphaned components, and file discovery issues. All findings become "wire it in" issues — never deletion issues.
/dead-code-audit — Codebase Unwired Code Audit
POLICY: Code is never deleted. It was created for a reason. Every finding becomes a "wire it in" issue.
Performs a comprehensive scan of the AutoBot codebase to find:
- Not wired — complete, implemented code that exists but isn't connected to the app
- Intentionally isolated — test fixtures, examples, stubs (marked SKIP)
There is no "DEAD" classification. Code with zero references is still NOT_WIRED — find the right place to connect it.
Findings are classified, deduplicated against existing issues, and batch-filed as GitHub issues for human review. No automatic deletion or wiring.
Usage
/dead-code-audit
Scans the entire codebase (no arguments). Runs 3 audit agents in parallel, classifies findings, and files issues.
Workflow
Phase 0: Tracker Cross-Reference Pre-flight (REQUIRED)
Before launching agents, run the automated tracker-cross-reference audit. This catches a class of findings the structural agents miss: production modules whose docstring cites a CLOSED tracker issue but have zero production callers. This is the systemic premature-closure pattern surfaced by #6836.
python3 pipeline-scripts/audit-unwired-trackers.py --json > /tmp/tracker-findings.json
jq 'length' /tmp/tracker-findings.json
Each entry in the output represents a feature whose tracker was closed before the integration step. Treat each as a NOT_WIRED finding for Phase 2 below.
The same script runs weekly via .github/workflows/unwired-tracker-audit.yml and files capped batches automatically — but running it interactively at audit time gives full visibility.
Phase 1: Parallel Audit Agents
Three agents run simultaneously, each scanning a different domain:
Agent A: Backend — Unregistered API Routers
Task: Find all Python files in autobot-backend/api/ that define an APIRouter but are not registered in any router registry.
Scan steps:
-
Find all files defining routers:
grep -rl "router = APIRouter" autobot-backend/api/ --include="*.py" -
Find all router files actually imported in registries:
grep -rh "from api\." autobot-backend/initialization/router_registry/ --include="*.py" | sort -u grep -rh "import.*api\." autobot-backend/initialization/router_registry/ --include="*.py" | sort -u -
Compare: files in step 1 but not referenced in step 2 = unregistered
-
For each unregistered router:
- Read the file and extract endpoint definitions
- Check if it's imported anywhere else (tests, docstrings)
- Check if there's already a GitHub issue for it (
gh issue list --search "<filename>") - Classify:
- Has complete endpoints + never imported anywhere →
NOT_WIRED - Has only placeholder/empty endpoints →
DEAD - Imported only in tests →
SKIP(test fixtures are intentional)
- Has complete endpoints + never imported anywhere →
-
For NOT_WIRED findings, extract:
- Endpoints defined (path + method)
- Module name and prefix
- Any git blame to understand why it wasn't wired
Report back with:
ROUTERS_FOUND: <count>
NOT_WIRED_ROUTERS: <list of filenames>
DEAD_ROUTERS: <list of filenames>
SKIPPED_ROUTERS: <list of filenames>
Per router:
<filename>: <count> endpoints, <endpoint_list>, classified as <NOT_WIRED|DEAD|SKIP>
Agent B: Backend — Uninvoked HookPoints + Unregistered Extensions
Task: Find HookPoint enum values that are defined but never emitted, and extensions that are defined but never registered.
HookPoint scan:
-
Get all defined HookPoints:
grep " [A-Z_]*" autobot-backend/extensions/hooks.py | sed 's/[[:space:]]*\([A-Z_]*\).*/\1/' -
Find actual invocation sites (where hooks are emitted):
grep -rn "invoke_hook\|\.invoke(" autobot-backend/ \ --include="*.py" \ --exclude="hooks.py" \ --exclude="manager.py" \ --exclude="*_test.py" -
Extract which HookPoints are actually invoked from step 2
-
Compare: HookPoints in step 1 but not in step 2 = uninvoked
-
Classify all uninvoked HookPoints as
NOT_WIRED(they were defined with intent, never used)
Extension registration scan:
-
List all extensions in builtin:
ls -1 autobot-backend/extensions/builtin/ | grep -E "\.py$" | sed 's/\.py$//' -
Find which extensions are registered at startup:
grep -rn "register_extension\|add_extension\|ExtensionManager" autobot-backend/initialization/ \ --include="*.py" -
Compare: extensions in builtin/ but not registered =
NOT_WIRED -
For each NOT_WIRED extension:
- Read class definition
- Check if it's referenced anywhere (tests, docs, plugins)
- Classify as
NOT_WIREDorSKIPif only in tests
Report back with:
HOOKPOINTS_TOTAL: <count>
HOOKPOINTS_INVOKED: <count> (e.g., 3 of 24)
HOOKPOINTS_NOT_WIRED: <list of names>
EXTENSIONS_IN_BUILTIN: <list>
EXTENSIONS_REGISTERED: <list>
EXTENSIONS_NOT_WIRED: <list>
Per hookpoint:
<NAME>: defined in hooks.py, 0 invocation sites, classified as NOT_WIRED
Per extension:
<ClassName>: defined in <file>, not registered at startup, classified as NOT_WIRED
Agent C: Frontend — Orphaned Views + Components + Shadow Files
Task: Find Vue views not in the router, orphaned components, and duplicate/shadow router files.
Orphaned views scan:
-
List all Vue files in views directory:
find autobot-frontend/src/views -name "*.vue" | sed 's|.*/||;s|\.vue||' -
Find all component registrations in routers:
cat autobot-frontend/src/router/index.ts | grep -oE "[A-Z][a-zA-Z]*View" cat autobot-frontend/src/App.vue | grep -oE "[A-Z][a-zA-Z]*View" -
Compare: views in step