Agent Skills: Gap Detection

Gap Detection

Overview

This skill performs a structured health-check scan on session start (or on-demand) to surface:

1. Files missing README / documentation headers
2. Test coverage gaps (source files without corresponding test files)
3. TODO / FIXME counts and locations
4. Undocumented public APIs and exports

Use this skill before beginning any significant work in an unfamiliar codebase, or as a recurring quality gate.

When to Use

• Session start in a new or unfamiliar repository
• Before planning a feature to understand existing quality debt
• During code review to identify undocumented additions
• As part of a proactive-audit pipeline

The Iron Law

NO HEALTH REPORT WITHOUT EVIDENCE — EVERY FINDING MUST CITE FILE AND LINE

Never report a gap without a concrete file path and (when applicable) line number. Vague summaries are not actionable.

Workflow

Step 1: Scan for undocumented source files

Command:

# Find source files without a README at their directory level
find . -type f \( -name "*.ts" -o -name "*.js" -o -name "*.cjs" -o -name "*.mjs" -o -name "*.py" \) \
  ! -path "*/node_modules/*" ! -path "*/.git/*" ! -path "*/dist/*" ! -path "*/build/*" \
  -print | while read f; do
    dir=$(dirname "$f")
    if [ ! -f "$dir/README.md" ] && [ ! -f "$dir/index.md" ]; then
      echo "NO_README: $f"
    fi
  done | sort -u

Expected output: Lines of form NO_README: ./src/utils/helper.ts

Verify: Exit code 0; non-empty output means gaps exist.

Step 2: Find test coverage gaps

Command:

# Find source files that have no matching test file
find . -type f \( -name "*.ts" -o -name "*.js" -o -name "*.cjs" \) \
  ! -path "*/node_modules/*" ! -path "*/.git/*" ! -path "*/dist/*" \
  ! -path "*/tests/*" ! -path "*/__tests__/*" ! -path "*.test.*" ! -path "*.spec.*" \
  -print | while read src; do
    base=$(basename "$src" | sed 's/\.[^.]*$//')
    dir=$(dirname "$src")
    if ! find . -path "*/tests/*" -name "${base}.test.*" 2>/dev/null | grep -q .; then
      if ! find . -name "${base}.test.*" 2>/dev/null | grep -q .; then
        echo "NO_TEST: $src"
      fi
    fi
  done

Expected output: Lines of form NO_TEST: ./src/auth/jwt.ts

Verify: Exit code 0; any NO_TEST lines are coverage gaps.

Step 3: Count and locate TODO/FIXME markers

Command:

grep -rn --include="*.ts" --include="*.js" --include="*.cjs" --include="*.mjs" --include="*.py" \
  -E "(TODO|FIXME|HACK|XXX):" \
  --exclude-dir=node_modules --exclude-dir=.git --exclude-dir=dist \
  . 2>/dev/null | sort

Expected output: ./src/auth/jwt.ts:42: // TODO: add token rotation

Verify: Total count printed via | wc -l appended to output.

Step 4: Detect undocumented public exports

Command:

# Find exported functions/classes without JSDoc or inline comment above them
grep -rn --include="*.ts" --include="*.js" -E "^export (function|class|const|async function)" \
  --exclude-dir=node_modules --exclude-dir=dist --exclude-dir=.git \
  . 2>/dev/null | while IFS=: read file line content; do
    prevline=$((line - 1))
    comment=$(sed -n "${prevline}p" "$file" 2>/dev/null | grep -E "(/\*|\*/|//)" || true)
    if [ -z "$comment" ]; then
      echo "NO_DOC: $file:$line $content"
    fi
  done

Expected output: NO_DOC: ./src/api/router.ts:15 export function handleRequest

Verify: Exit code 0.

Step 5: Compile health report

After running Steps 1–4, produce a structured report with:

• Summary table: counts per category (NO_README, NO_TEST, TODO/FIXME, NO_DOC)
• Top 10 highest-priority gaps (ranked by: public API > module entrypoints > internals)
• Recommended next actions (e.g., "Add README to src/auth/, add tests for jwt.ts")

Report format:

## Gap Detection Report — {{date}}

| Category   | Count |
| ---------- | ----- |
| NO_README  | {{n}} |
| NO_TEST    | {{n}} |
| TODO/FIXME | {{n}} |
| NO_DOC     | {{n}} |

### Priority Gaps

1. {{file:line}} — {{reason}}
   ...

### Recommended Actions

- [ ] {{action}}
      ...

Verify: Report written to .claude/context/tmp/gap-detection-report-{{date}}.md.

Enforcement Hooks

Input validated against schemas/input.schema.json before execution. Output contract defined in schemas/output.schema.json.

Memory Protocol

Before starting:

Read .claude/context/memory/learnings.md for prior gap scans and known chronic issues.

After completing:

• Append gap summary to .claude/context/memory/learnings.md with date and repo path
• If critical gaps found (0% test coverage on a module, zero README in public API dir), append to .claude/context/memory/issues.md

Anti-Patterns

• Never report "no gaps found" without actually running the scan commands
• Never produce a gap report without file paths — vague summaries are unusable
• Never suppress TODO/FIXME output — they represent deferred debt
• Never run on node_modules/, dist/, .git/ directories

Related Skills

• proactive-audit — broader audit including hook syntax and agent consistency
• tdd — use after gap-detection to address test coverage gaps
• debugging — follow-up for runtime gaps identified during scan
• context-compressor — compress large gap reports before handing off

Assigned Agents

• developer (primary — runs on feature work)
• qa (primary — runs before test strategy)
• architect (supporting — runs before architecture reviews)
• planner (supporting — runs before planning sessions)