/cheat-status — Status Dashboard

Overview

[User: status]
  ↓
[Phase 1: Read .cheat-state.json + scan file system]
  ↓
[Phase 2: Calculate derived metrics]
  ↓
[Phase 3: Detect suggestion triggers (upgrade / bump / cleanup)]
  ↓
[Phase 4: Output dashboard]

Constants

• SQLITE_UPGRADE_THRESHOLD = 30 — Suggest SQLite upgrade when calibration_samples reaches N
• CLEANUP_LINE_THRESHOLD = 600 — Suggest cleanup when rubric_notes.md exceeds N lines
• STALE_PREDICTION_DAYS = 30 — Prompt cleanup for in_progress predictions not published for over N days

Inputs

.cheat-state.json

predictions/*.md

candidates.md

rubric_notes.md

.cheat-cache/usage.jsonl

Workflow

Phase 1: Read State

state = read_json('.cheat-state.json')
if not state:
    return "You haven't initialized yet. Please run /cheat-init first."

predictions = glob('predictions/*.md')
candidates_count = parse_candidates_md_entries()
rubric_lines = wc -l rubric_notes.md

Phase 2: Derived Metrics

len(state.shoots)

buffer_days = buffer_count × target_publish_cadence_days

<1 Red / 1-2 Orange / 3-5 Green / >5 Blue

target_publish_cadence_days=null

calibration_samples

now - state.shoots[0].shot_at

Phase 3: Detect Suggestion Triggers

1. Buffer Color = 🔴 Red → High-priority alert in first line: "Buffer has 0/1 piece(s), risk of breaking schedule on next publish date — you must shoot ≥1 piece today. Say 'recommend topics' and I'll only push top 1 safe-scoring (non-experimental) option"
2. Buffer Color = 🔵 Blue → High-priority prompt: "Buffer has N backlogged pieces. Pause shooting, publish existing stock + do retros first. Say 'published ...' and I'll dequeue it for you"
3. Earliest shot_at in state.shoots > 14 days → "You have a video shot N days ago but not published — risk of topic timeliness loss, recommend publishing or discarding as soon as possible"
4. in_progress is stale (>= STALE_PREDICTION_DAYS) → High-priority prompt "Clean up or publish"
5. Pending retros ≥ 1 → High-priority "You need to retro X pieces today"

6. pool_status=none

   +

   calibration_samples=0

   + >24h since init → "🌱 It's been N days since you initialized but you haven't shot anything — is it because you have no topics? Run /cheat-seed to get 5 candidates + 5 drafts in 5 minutes" (high priority)
7. Claude detects systematic bias signals (not strictly ≥3 consecutive same-direction) → Prompt "Recommend running /cheat-bump"
   • Default reference: ≥3 consecutive same-direction deviations
   • But Claude can trigger earlier: 1 extreme deviation (≥10x) or 2 same-direction + strong reverse evidence in comments
   • Or trigger later: 3 same-direction but each magnitude <25% (may just be noise)
   • Explicitly label when prompting: "This is [default-aligned] / [judgment-driven]"
8. calibration_samples enters new confidence level (0→1, 2→3, 5→6, 10→11, 20→21) → Prompt "🎉 Confidence upgraded: <old level> → <new level>. Bucket center precision improved from ±X% to ±Y%". Notification only, no mandatory user action — all skills automatically adjust based on calibration_samples
9. calibration_samples crosses 5 → "Your rubric is ready for its first official bump. Review rubric_notes.md to see if there are patterns supported by ≥3 samples → run /cheat-bump"
10. calibration_samples crosses 10 → "You can run /cheat-bump --bucket-only --scheme percentile to switch bucket boundaries to percentile (always self-consistent)"
11. calibration_samples crosses SQLITE_UPGRADE_THRESHOLD and data_layer=markdown → "Recommend running tools/md-to-sqlite.py"
12. rubric_notes.md lines > CLEANUP_LINE_THRESHOLD → "Recommend cleaning up observation sections (manually or triggered on next bump)"
13. calibration_samples ≥ 5 + pool_status=none → "You can start building a topic pool"
14. calibration_samples ≥ 15 + pool_status=none → "Strongly recommend building a pool: run /cheat-trends or manually create candidates.md"
15. state.hooks_installed=false → "Your immutability is based on gentleman's agreement, recommend running /cheat-init to install hooks"
16. state.last_bump_self_audited=true → "Last bump was self-audited. Recommend configuring mcp__llm-chat__chat for external review on next bump"
17. state.rubric_form_mismatch=true → "Your content_form is not opinion-video, but you're using the built-in opinion rubric. Early predictions will be less accurate, recommend adjusting weights to fit your format on next bump"
18. state.benchmark_status=pending → "🎯 You promised to find benchmark accounts during init but haven't done so yet. Run /cheat-learn-from to import ≥3 benchmark videos, and the tool will have an anchor"
19. state.benchmark_status=imported + Claude judges user data signals have exceeded benchmark → "📊 Your actual data has become the main signal, benchmark influence is fading"

• Default reference: calibration_samples ≥ 10
• But Claude can trigger earlier: N=5 but ≥3 pairs of (score, actual performance) in user samples are inconsistent with benchmark patterns — indicating your account has deviated from the benchmark path
• Or trigger later: N=15 but user samples are very similar with insufficient diversity → benchmark still has signal value
• Prompt is notification not gate — benchmark.md is always retained for sanity check, cheat-seed can still read it

Phase 4: Output Dashboard

🎛️ cheat-on-content Status (Updated on 2026-05-04 15:00)

Content Format: opinion-video / Duration 3-5min / Cadence: Every other day
Current rubric: v2 (Last bump: 2026-04-22)
Calibration Samples: 18 pieces
Confidence: 🟢 High (Center ±15%, rubric format stable)
Baseline: 42k median views

📦 Buffer: 3 pieces (🟢 Green)
   Based on your cadence (every other day) = 6-day buffer, rhythm is stable

📊 Progress Bars
  [█████████████░░░░░] 18 / 30 → SQLite upgrade recommendation threshold
  [██████████░░░░░░░░] 18 / 10 → Percentile buckets available (threshold exceeded)

🎬 To-Do (By Urgency)
  🚨 Retro 1 piece (Past T+3d)
     - predictions/2026-05-01_db063817_You're No Longer in the Relationship.md (T+3d reached)
  ⚠️  3 consecutive same-direction deviations (high, high, high) → Recommend /cheat-bump
  💤 in-progress prediction has been stale for 35 days
     - predictions/2026-04-01_xxx.md → Did you publish it and forget to record? Or discard it?

🔥 Candidate Pool
  - candidates.md: 27 entries (tier1: 12, tier2: 9, tier3: 6)
  - Days since last hot topic capture: 4 days — You can run /cheat-trends again

📈 Health Status
  - rubric_notes.md: 412 lines (Healthy, <600 warning line)
  - hooks_installed: ✅
  - external audit configured: ❌ → Recommend configuring mcp__llm-chat__chat

Next Steps (By Recommendation Priority):
1. /cheat-retro predictions/2026-05-01_db063817_You're No Longer in the Relationship.md  ← Most urgent
2. /cheat-bump  ← Handling 3 consecutive same-direction deviations
3. Process stale in-progress (manually or reply "clean up in-progress")

Complete command list can be found in the main SKILL.md.

Key Rules

1. No side effects. Read-only, no writes. Any state modification is handled by other skills
2. Don't pretend data is available. If state file fields are missing → explicitly mark "Unknown", don't guess
3. Suggestions with priority. Users will be overwhelmed if 10 suggestions are shown at once — sort by urgency
4. Attach commands to each suggestion. Don't just say "Time to bump" — provide exact entry like

   /cheat-bump --propose "..."

Refusals

• "By the way, help me run retro automatically" → Refuse. Status is read-only, retro is a separate action (avoid doing two things in one operation)
• "I don't want to see rubric_notes line count, it's too trivial" → Still include it but collapse to the bottom "Health Status" section — presence of status information lets users see issues before they occur

Integration

• Upstream: All other skills update .cheat-state.json upon completion, status is the visualization of these updates
• Downstream: Each suggestion routes to a specific sub-skill
• Meta-logging hook (if enabled) → Writes to usage.jsonl, status uses it to calculate "number of times since last X"