adev-codehealth

name

adev:codehealth

description

Proactively scan source code for dead exports, orphan files, unused dependencies, stale code, and duplicate logic. Produces severity-tiered reports from repomap artifacts. Use when the user says 'check code health', 'find dead code', 'unused exports', 'stale files', 'orphan files', 'unused dependencies', or wants to identify cleanup opportunities before refactoring. In OpenCode, invoke with skill({ name: 'adev:codehealth' })

Code Health Scanner

Scan source code for dead exports, orphan files, unused dependencies, stale code, and duplicate logic. Produces a severity-tiered markdown report from /adev:repomap artifacts.

Announce at start: "I'm using the adev:codehealth skill to scan for code health issues."

Arguments

Argument	Required	Description
(none)	—	Full scan: all passes against all hygiene.source_roots
--module <slug>	No	Restrict scan to a single manifest module's paths
--check <name>	No	Comma-separated check filter. Valid: dead-exports, orphan-files, unused-deps, stale-code, duplicate-logic

Both --module and --check may be combined (intersection: only named checks, only files in the module).

Prerequisites

Complete these checks in order. Stop at the first failure.

1. Manifest Validation

Read .context-index/manifest.yaml. If missing or unreadable:

INVALID_MANIFEST: Missing or invalid manifest.yaml — source_roots is required for codehealth scanning.

Verify hygiene.source_roots exists and is a non-empty list. If absent:

INVALID_MANIFEST: manifest.yaml is missing hygiene.source_roots — required for codehealth scanning.

2. Repomap Artifact Validation

Check that both files exist:

• .context-index/hygiene/symbol-ranks.json
• .context-index/hygiene/dependency-graph.json

If either is missing:

MISSING_REPOMAP: Repomap artifacts not found. Run /adev:repomap first to generate the symbol index and dependency graph.

Stop without running any passes.

3. Argument Validation

If --module <slug> is provided, resolve against manifest.yaml modules[].slug. If not found:

UNKNOWN_MODULE: Unknown module '<slug>'. Available modules: <comma-separated list of slugs>.

If --check <name> is provided, validate each name against the allowlist: dead-exports, orphan-files, unused-deps, stale-code, duplicate-logic. If any is unrecognized:

UNKNOWN_CHECK: Unknown pass '<name>'. Valid passes: dead-exports, orphan-files, unused-deps, stale-code, duplicate-logic.

4. File Scope Resolution

Resolve the set of files to scan:

1. Load hygiene.source_roots from manifest (e.g., ["cli/", "hooks/", "skills/", "templates/"])
2. If --module <slug> provided: intersect with the module's paths from manifest.yaml modules[]
3. Subtract hygiene.coverage_exclude glob patterns (e.g., ["tests/**", "skills/*/evals/**"])
4. The resulting file list is the scan scope for all passes

Load Skill Extensions: Load any skill extension instructions before proceeding:

adev skill-ext load --skill codehealth

If the output is not __NONE__, incorporate it as additional standing instructions that apply to this skill's entire execution. Frame it as: "The following skill extension instructions apply to this invocation (source: installed domain extensions and/or project-level overrides)." If the output is __NONE__, continue normally.

---

Detection Passes

Execute passes in this fixed order. If --check is provided, skip passes not in the filter.

Scope types: Passes 1 and 2 are dependency-graph-scoped (analyze only files present in dependency-graph.json nodes). Passes 3 and 4 are file-system-scoped (scan all files within resolved source_roots). Pass 5 requires tree-sitter at runtime.

Pass 1: Dead Export Detection (dead-exports)

Goal: Find exported symbols with zero imports across the codebase.

Steps:

1. Read .context-index/hygiene/dependency-graph.json. Parse nodes[] and edges[]. If the JSON has unexpected format, skip this pass with note:

   FORMAT_ERROR: dependency-graph.json format error — pass skipped.

2. For each node in nodes[] within the resolved file scope:

   • For each symbol in the node's exports[]:
     • Search all edges[] for an edge where edge.to matches the node's path AND the symbol appears in edge.symbols[].
     • If no such edge exists, the symbol is a dead-export candidate.
   • Suppression by non-code references (rev-2 schema): If the node carries tags including "public-api-entry", OR if symbol-ranks.json records referenceSources[] containing "package-exports" or "doc-reference" for this symbol, the symbol is NOT dead — skip it. The new referenceSources[] field encodes inbound references that the static .mjs-only edge analysis cannot see (SKILL.md prose mentions; package.json:exports entries).

3. Cross-reference with .context-index/hygiene/symbol-ranks.json to get line number for each dead export. If symbol-ranks.json has unexpected format, skip line number enrichment with note:

   FORMAT_ERROR: symbol-ranks.json format error — line numbers unavailable.

4. Severity classification:

   • high: The dead export is the only export in its file (entire file may be dead code).
   • medium: The dead export coexists with other referenced exports in the same file.
   • low: The symbol's file appears as the to target of an edge with type: "re-export" (barrel file re-export — may be consumed externally outside the project).

   Tolerance contract: Unknown edge.type values introduced by future schema evolutions (e.g., the rev-2 "doc-reference" type) MUST NOT cause this pass to crash; treat unknown types as informational and continue.

5. Emit findings: { pass: "dead-exports", severity, file_path, line_number, symbol, description }.

Pass 2: Orphan File Detection (orphan-files)

Goal: Find source files not imported by any other file.

Steps:

1. From dependency-graph.json, collect all file paths appearing as edge.to in any edge. These are "imported files." (Note: doc-reference edges introduced in rev-2 of core-parser-pipeline.spec.md also populate edge.to, so a file referenced from a SKILL.md is counted here and will not be flagged as an orphan.)
2. For each node in nodes[] within the resolved file scope:
   • If the node's path does NOT appear in the "imported files" set, it is an orphan candidate.
3. Exclude entry points from orphan detection:
   • Files matching **/index.* (barrel files / entry points)
   • Files matching **/cli.* or **/main.* (CLI entry points)
   • Files listed as hook scripts in hooks/hooks.json
   • Files whose FileNode tags includes "public-api-entry" (rev-2 schema — these files are declared as the package's public API surface via package.json:exports and are consumed by external dependents not visible in the local graph).
   • Test files matching hygiene.coverage_exclude patterns
4. Verify candidates against hook scripts: Before reporting an orphan, grep hook shell scripts (hooks/*.sh) for the module's filename or its exported symbol names. Hooks often contain inline Node.js blocks with dynamic import() calls that reference library modules — these are invisible in the static .mjs dependency graph but are real consumers.
5. Severity classification:
   • high: The orphan file has no outgoing edges either (imports nothing and is imported by nothing — fully isolated).
   • medium: The orphan file has outgoing edges (imports others but nobody imports it — possible unused entry point).
6. Emit findings: { pass: "orphan-files", severity, file_path, description }.

Pass 3: Unused Dependency Detection (unused-deps)

Goal: Compare package.json dependencies against actual imports in source files.

Steps:

1. Read package.json from the project root. If missing, skip this pass with note:

   MISSING_PACKAGE_JSON: package.json not found — unused dependency detection skipped.

2. Collect all package names from dependencies and devDependencies.

3. Scan all source files within the resolved source_roots (not just dependency graph nodes) for import patterns:

   • import ... from '<package>' or import ... from '<package>/...'
   • import('<package>') or import('<package>/...')
   • require('<package>') or require('<package>/...') (included for consumer project compatibility — this project is pure ESM per constitution)

4. For each package not matched by any import pattern:

   • Severity:
     • high: Package is in dependencies (production dependency, definitely unused).
     • medium: Package is in devDependencies (may be used by scripts, tooling, or test runners not captured by import analysis).

5. Emit findings: { pass: "unused-deps", severity, file_path: "package.json", symbol: "<package-name>", description }.

Pass 4: Stale Code Detection (stale-code)

Goal: Flag files not modified recently relative to their module's activity.

Steps:

1. Read hygiene.staleness_threshold_days from manifest (default: 30 if absent).

2. For each source file within the resolved scope, run:

   git log -1 --format=%ci -- <file-path>
   

   Extract the last commit date. If git log fails or git is unavailable, skip this pass with note:

   GIT_UNAVAILABLE: git not available or git log failed — stale code detection skipped.

3. Group files by module (from manifest.yaml modules[].paths).

4. For each module, find the most recent commit date across all its files.

5. For each file whose last commit date is older than staleness_threshold_days ago:

   • Skip if uniformly old: If ALL files in the module have the same "most recent" date (the entire module is uniformly aged), emit no staleness findings for that module.
   • Otherwise, check if the file also appears as a dead export or orphan finding from earlier passes.

6. Severity classification:

   • high: The stale file also has zero references (dead export or orphan finding exists for this file).
   • low: The stale file is still actively referenced (stable code, not necessarily problematic).

7. Emit findings: { pass: "stale-code", severity, file_path, description: "Last modified: <date>. Module last active: <date>." }.

Pass 5: Duplicate Logic Detection (duplicate-logic)

Goal: Find structurally similar or identical function bodies across files.

Prerequisite: Check if tree-sitter is available at runtime (use the same detection mechanism as /adev:repomap — check for web-tree-sitter module availability). If tree-sitter is NOT available:

TREESITTER_UNAVAILABLE: Duplicate logic detection skipped — tree-sitter not available. Run /adev:repomap with tree-sitter enabled for this pass. (Per ADR-0001, tree-sitter is optional.)

Skip this pass entirely.

Steps (if tree-sitter available):

1. For each source file within the resolved scope, parse with tree-sitter to extract function/method AST nodes.
2. For each pair of functions across different files, compare structural similarity:
   • Normalize variable names (replace with positional placeholders).
   • Compare AST node types and structure.
3. Severity classification:
   • high: Exact duplicates (identical function bodies after normalization).
   • medium: Structurally similar (same AST shape, different variable names or literal values).
4. Emit findings: { pass: "duplicate-logic", severity, file_path, line_number, symbol, description: "Duplicate of <other-file>:<line> (<symbol>)" }.

---

Severity Reference

Pass	High	Medium	Low
dead-exports	Only export in file	Coexists with referenced exports	Re-export target
orphan-files	Fully isolated (no edges)	Has outgoing edges only	—
unused-deps	dependencies package	devDependencies package	—
stale-code	Stale + unreferenced	—	Stale but referenced
duplicate-logic	Exact duplicate	Structural similarity	—

---

Report Generation

After all passes complete, generate the report.

Report Path

.context-index/reports/codehealth-<YYYY-MM-DD>.md

Create the .context-index/reports/ directory if it does not exist. If a report for today's date already exists, overwrite it (idempotent).

If unable to write:

WRITE_ERROR: Unable to write report — check directory permissions.

Report Format

Persona adaptation: The report written to disk always uses the full format below. The chat summary presented to the user should follow the active persona's output rules.

---
date: <ISO 8601 timestamp>
module_filter: <--module value or "all">
check_filter: <--pass value or "all">
total_findings: <count>
summary:
  high: <count>
  medium: <count>
  low: <count>
---

Summary Table

# Code Health Report

| Pass | High | Medium | Low | Total |
|------|------|--------|-----|-------|
| Dead Export Detection | N | N | N | N |
| Orphan File Detection | N | N | N | N |
| Unused Dependency Detection | N | N | N | N |
| Stale Code Detection | N | N | N | N |
| Duplicate Logic Detection | N | N | N | N |
| **Total** | **N** | **N** | **N** | **N** |

Per-Pass Sections

For each pass that ran:

If findings exist:

## Dead Export Detection

| Severity | File | Line | Symbol | Description |
|----------|------|------|--------|-------------|
| high | src/utils/old.mjs | 12 | unusedHelper | Only export in file, zero references |
| medium | src/lib/api.mjs | 45 | legacyFetch | Coexists with 3 referenced exports |

Sort findings by severity (high first), then file path ascending, then line number ascending. Show — for optional fields (Line, Symbol) when absent.

If zero findings:

## Dead Export Detection

No issues found.

If pass was skipped:

## Duplicate Logic Detection

Skipped — tree-sitter not available. Run `/adev:repomap` with tree-sitter enabled.

Malformed Findings

If any finding is missing required fields (pass, severity, file_path, description), skip it and note in the report:

MALFORMED_FINDING: N finding(s) omitted due to missing data.

Empty Results

If ALL passes produced zero findings:

# Code Health Report

**No code health issues found.**

| Pass | High | Medium | Low | Total |
|------|------|--------|-----|-------|
| Dead Export Detection | 0 | 0 | 0 | 0 |
| ... | ... | ... | ... | ... |
| **Total** | **0** | **0** | **0** | **0** |

Conversation Summary

After writing the report, print to the conversation:

Code health scan complete.

Findings: N high, N medium, N low (N total)

Top issues:
1. [high] src/utils/old.mjs:12 — unusedHelper: Only export in file, zero references
2. [high] package.json — lodash: Production dependency not imported anywhere
3. [medium] src/lib/api.mjs:45 — legacyFetch: Coexists with 3 referenced exports

Full report: .context-index/reports/codehealth-<date>.md

Show top 3 findings sorted by severity descending, then file path ascending. If zero findings:

Code health scan complete. No issues found.

Full report: .context-index/reports/codehealth-<date>.md