Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Diagnose and repair a broken or non-standard agent-second-brain install/migration. Triggers — upgrade.sh failed, bot won't start after upgrade, "миграция сломалась", "бот не стартует после обновления", legacy d-brain units still present, brain session won't boot, doctor reports failures.
.claude/skills/smixs-migrate-doctor/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 1% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 13% | 0% |
You are running INSIDE an interactive Claude Code session on the user's server (subscription billing — this is the supported way to use AI for repair). Your job: get a broken or half-migrated install to a healthy v3.0 state. The deterministic path is bash upgrade.sh — your job is to diagnose WHY it fails, remove the obstacle, and re-run it. Do not re-implement the migration by hand unless upgrade.sh itself cannot work.
| Version | Markers | Notes | |---|---|---| | v1 (Dec 2025) | system-level /etc/systemd/system/d-brain-bot.service, TODOIST_API_KEY in .env, mcp-config.json | Todoist/MCP era, claude -p per message | | v2 | d-brain-* units, claude -p/claude --print pipeline, weekly timer | headless calls — dead after 2026-06-15 billing change | | v3.0 (target) | dbrain-* systemd --user units, persistent tmux brain, ~/.dbrain/ runtime dir, cron subsystem | interactive session on subscription |
bash# What's installed and running? systemctl --user list-units 'dbrain-*' --all sudo systemctl list-units 'd-brain-*' --all # legacy system-level ls ~/.dbrain/ 2>/dev/null # runtime dir (v3) tmux ls 2>/dev/null # brain sessions git -C ~/projects/agent-second-brain log --oneline -3 cat ~/projects/agent-second-brain/.env | grep -v 'TOKEN\|KEY' # never print secrets claude auth status --json # needs "loggedIn": true journalctl --user -u dbrain-bot -n 50 --no-pager
tar czf ~/dbrain-backup-$(date +%s).tgz -C ~/projects agent-second-brain --exclude=.venv and note the current commit (git rev-parse HEAD).vault/daily, notes, cards). The vault is the user's data; migration touches code, units and runtime files only.bash upgrade.sh after each fix — it is idempotent. Fix the obstacle, not the symptom.git pull --ff-only → git stash (show the user what was stashed)sudo systemctl disable --now 'd-brain-*' and remove files from /etc/systemd/system/loggedIn: false → tell the user to run claude interactively and log in; do NOT script OAuthloginctl enable-linger $USERTODOIST_API_KEY / DBRAIN_MODE lines in .env → remove them (v3 ignores but they confuse humans)dbrain restart or tmux kill-session -t <brain> (the bot recreates it lazily)bashbash scripts/check-no-claude-p.sh # guard clean systemctl --user is-active dbrain-bot dbrain-watchdog dbrain-process.timer dbrain-doctor.timer uv run python -m d_brain.services.doctor # ok=True cat ~/.dbrain/STATUS.md # state: healthy
Finish with a short summary: what was broken, what you changed, what the user should watch for. Suggest sending the bot a test voice message.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,286 | 3,875 | -10% | 1 | 1 | 0% | 195 | 1,423 | +630% | 0 | 0 | — |
case-02 | fail→fail | 5,563 | 10,004 | +80% | 1 | 1 | 0% | 310 | 1,355 | +337% | 0 | 0 | — |
case-03 | fail→fail | 4,663 | 2,953 | -37% | 1 | 1 | 0% | 151 | 1,197 | +693% | 0 | 0 | — |
case-04 | fail→fail | 2,771 | 3,639 | +31% | 1 | 1 | 0% | 303 | 1,365 | +350% | 0 | 0 | — |
case-05 | pass→pass | 9,015 | 4,703 | -48% | 1 | 1 | 0% | 1,535 | 1,543 | +1% | 0 | 0 | — |
case-06 | pass→pass | 7,645 | 4,014 | -47% | 1 | 1 | 0% | 1,460 | 1,653 | +13% | 0 | 0 | — |
case-07 | pass→pass | 12,208 | 3,821 | -69% | 1 | 1 | 0% | 2,165 | 1,554 | -28% | 0 | 0 | — |
case-08 | pass→pass | 10,046 | 3,294 | -67% | 1 | 1 | 0% | 1,556 | 1,366 | -12% | 0 | 0 | — |
case-09 | pass→pass | 4,149 | 3,551 | -14% | 1 | 1 | 0% | 705 | 1,371 | +94% | 0 | 0 | — |
case-10 | fail→fail | 8,001 | 3,026 | -62% | 1 | 1 | 0% | 1,315 | 1,220 | -7% | 0 | 0 | — |
case-11 | pass→pass | 3,384 | 2,991 | -12% | 1 | 1 | 0% | 479 | 1,355 | +183% | 0 | 0 | — |
case-12 | fail→fail | 8,478 | 1,969 | -77% | 1 | 1 | 0% | 1,504 | 1,174 | -22% | 0 | 0 | — |
case-13 | fail→pass | 6,366 | 2,460 | -61% | 1 | 1 | 0% | 993 | 1,291 | +30% | 0 | 0 | — |
case-14 | fail→pass | 11,439 | 2,445 | -79% | 1 | 1 | 0% | 1,795 | 1,222 | -32% | 0 | 0 | — |
case-15 | fail→fail | 7,421 | 5,439 | -27% | 1 | 1 | 0% | 1,359 | 1,470 | +8% | 0 | 0 | — |
case-16 | fail→pass | 7,169 | 3,545 | -51% | 1 | 1 | 0% | 1,207 | 1,450 | +20% | 0 | 0 | — |
case-17 | pass→pass | 8,334 | 1,552 | -81% | 1 | 1 | 0% | 1,351 | 1,137 | -16% | 0 | 0 | — |
case-18 | pass→pass | 9,954 | 1,602 | -84% | 1 | 1 | 0% | 1,606 | 1,085 | -32% | 0 | 0 | — |
case-19 | pass→pass | 7,977 | 3,197 | -60% | 1 | 1 | 0% | 1,325 | 1,437 | +8% | 0 | 0 | — |
case-20 | fail→fail | 25,378 | 6,688 | -74% | 1 | 1 | 0% | 5,936 | 1,295 | -78% | 0 | 0 | — |
case-21 | fail→fail | 12,252 | 7,417 | -39% | 1 | 1 | 0% | 2,338 | 1,174 | -50% | 0 | 0 | — |
case-22 | fail→fail | 12,064 | 8,242 | -32% | 1 | 1 | 0% | 2,318 | 1,396 | -40% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 18 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.