KevinZai $ccc-xray — Project Health Scorecard
Scan this repo and return a scorecard. Click-first UX — the user never types a number.
Promoted from commands/ccc-xray.md (legacy slash command).
Response shape (EVERY time)
Output these three sections in order.
1. Brand header
**CC Commander** · X-Ray Scanner · [Docs](https://commanderplugin.com)
2. Context strip (one line)
Run in parallel via a single Bash call:
git rev-parse --abbrev-ref HEAD 2>/dev/null→ branchgit ls-files | wc -l→ file count- Detect stack: presence of
package.json/pyproject.toml/Cargo.toml/go.mod
Render:
🔬 Target:
<branch>· files · stack: · ready to scan.
If not a git repo: "🔬 Not a git repo — pick a scan scope below, I'll run shallow checks."
3. Scope picker — AskUserQuestion
Call AskUserQuestion with 4 options (never list numbered options as text):
question: "What depth?"
header: "X-Ray Scan"
multiSelect: false
options:
- label: "⚡ Quick scan (5 areas)"
description: "Quality, docs, tests, deps, security. ~20s."
preview: "Recommended for most repos — trims perf + CI."
- label: "🔬 Full scan (7 areas)"
description: "Adds performance + CI. ~45s."
preview: "Use when prepping for a release or new-to-repo onboarding."
- label: "🔐 Focus: security"
description: "npm audit + secrets scan + dep age only."
preview: "Fastest path to 'am I leaking secrets?'"
- label: "📦 Focus: deps"
description: "Outdated deps, lockfile hygiene, duplicate packages."
preview: "Use before a dep bump PR."
Prepend ⭐ to Quick scan by default — it's the recommended starting point.
Native workflow (Claude Code v2.1.154+)
If the Workflow tool is available, prefer invoking the bundled audit workflow for the Full scan depth (and optionally Quick scan when you want adversarially verified findings):
Workflow({
scriptPath: "${CLAUDE_PLUGIN_ROOT}/workflows/ccc-audit.workflow.js",
args: { target: ".", dimensions: ["quality","security","perf","deps","tests","docs","ci"] }
})
Pass dimensions to match the user's chosen scope (e.g. ["security","deps"] for the focused security/deps picks). The workflow fans out one agent per dimension, adversarially verifies each finding, and returns a synthesized scorecard — false positives are filtered before they reach you.
If Workflow is unavailable (older client or tool not allowed), fall back to the Agent-based flow below — leave that flow intact.
Running the scan
After user picks, run checks in parallel via a single Bash call. Each check is cheap + silent-on-failure:
| Area | Check (Bash) |
|---|---|
| Quality | find . -type f \( -name '*.ts' -o -name '*.js' -o -name '*.py' \) -not -path './node_modules/*' | wc -l (size spread) |
| Docs | Test for README.md, CLAUDE.md, CHANGELOG.md presence |
| Tests | find . -path '*/tests/*' -o -path '*/__tests__/*' -o -name '*.test.*' | head -50 | wc -l |
| Deps | npm outdated --json 2>/dev/null || true (or pip list --outdated) |
| Security | npm audit --json 2>/dev/null || true (or pip-audit -f json) |
| Performance | bundle-size heuristic: du -sh dist 2>/dev/null + check for missing memoization patterns |
| CI | Presence of .github/workflows/, .gitlab-ci.yml, .circleci/ |
Translate each result to a 0-100 score with the rubric:
- 90-100: 🟢 healthy
- 70-89: 🟡 acceptable
- 50-69: 🟠 needs attention
- 0-49: 🔴 critical
Scorecard output (markdown table)
Render this shape:
### 🔬 X-Ray Report — <repo-name>
| Area | Score | Top finding | Fix |
|------|-------|-------------|-----|
| 🧹 Code quality | 🟢 88 | 2 files >800 lines | Spawn `$ccc-makeover` |
| 📚 Docs | 🟡 72 | CHANGELOG stale (14d) | Spawn docs-update task |
| 🧪 Tests | 🟠 54 | Coverage 41% | Run `/tdd` on missing modules |
| 📦 Deps | 🟡 70 | 4 minor updates pending | Run `npm outdated` |
| 🔐 Security | 🟢 91 | Clean | — |
| ⚡ Performance | 🟢 85 | Bundle 420 KB | — |
| 🚢 CI | 🔴 20 | No `.github/workflows/` | Spawn CI setup task |
**Overall: 66 · 🟠 needs attention · fix CI first.**
Post-scorecard: spawn fix offers
For every LOW-scoring row (🟠 or 🔴), append a "Fix it?" chip via mcp__ccd_session__spawn_task:
Example (pseudocode):
Spawn a background task to: "Set up GitHub Actions CI for this repo — test + lint + deploy. See
.github/workflows/conventions."
Each chip is ONE spawn_task call. Do NOT auto-execute — only offer. User clicks to spawn.
For 🟢/🟡 rows: no chip, just the score row.
Priority rule (single recommendation line)
At bottom, state which area to fix first based on impact × effort:
- If 🔴 Security → always first (blocks release)
- Else if 🔴 CI and repo has no deploy → CI first (unlocks everything)
- Else lowest score wins
Output: **🎯 My call: fix <area> first — <one-line rationale>.**
Argument handling
$ccc-xray quick→ skip picker, run Quick scan directly$ccc-xray full→ skip picker, run Full scan directly$ccc-xray security/deps/perf→ skip picker, run focused scan$ccc-xraybare → show the picker
Anti-patterns — DO NOT
- ❌ Run scans BEFORE user picks scope (wastes tokens on rejected depth)
- ❌ Render 7 rows when Quick scan was selected (show 5)
- ❌ Auto-spawn fix tasks without offering — chip pattern is click-first
- ❌ Output ASCII banners or box-drawing — plain markdown tables render everywhere
- ❌ Hardcode repo name — read from
git rev-parse --show-toplevelbasename - ❌ Block waiting for slow network audits — use
|| truefallbacks, never fail the whole scan
Brand rules
- Emoji-forward, scan-friendly (PM Consultant voice)
- Lead with the scorecard table, not narrative
- One recommendation line at the end, always
- ⭐ marks the recommended option in every picker
Cost analysis (NEW in beta.11+)
When the user asks for $ccc-xray cost or $ccc-xray full and a /cost transcript is available, parse it via the bundled utility at lib/cost-parser.js (ESM, side-effect-free):
import { parseCostOutput } from './lib/cost-parser.js';
const report = parseCostOutput(rawCostText);
// → { totalUsd, models, cacheHitRate, turns, avgUsdPerTurn, skipped, raw }
Use the result to render a "💰 Cost" row in the scorecard:
🟢ifcacheHitRate ≥ 0.5andavgUsdPerTurn < $0.10🟡ifcacheHitRate ∈ [0.2, 0.5)oravgUsdPerTurn ∈ [$0.10, $0.50)🟠ifcacheHitRate < 0.2oravgUsdPerTurn ≥ $0.50
The parser is liberal: malformed lines land in skipped[] (not thrown), empty input returns a zero-state object. Safe to call on any text the user pastes.
Tips for the agent executing this skill
- Whole flow is ≤3 turns: context+picker → user clicks → parallel scan → scorecard + chips.
- Run ALL checks in a single Bash call (chained with
;) to save turns. - If a check fails (network down, no npm), show the row as
⚪ n/a— don't crash. - Keep the scorecard table under ~15 rows total — summarize sub-findings in "Top finding" column.
Bottom line: pick scope → scan in parallel → one markdown table → one recommendation → chips for 🟠/🔴 rows. User clicks to fix.
⚙️ Fable contract: plan before build · verifier ≠ worker · prove before alarm · loops need gates · leave durable state —
rules/fable-method.md
(On Codex, present these options as a numbered list and ask the user to reply with a number — AskUserQuestion is Claude-only.)
(The Workflow(...) tool is not available on Codex — run the steps sequentially.)