Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Turn a REAL Slack thread (or channel) into a logged decision — read it, extract what was decided, who owns what, and record it in Notion — not a template for writing decisions. Use when asked to capture the decision from this thread, log what we agreed, turn this Slack discussion into a decision record, or close this out in Cowork. Reads the thread via the Slack connector, distils the decision / owners / next steps / open questions, and produces a decision-record artifact written to a Notion dat
.claude/skills/mohitagw15856-thread-to-decision-live/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 5 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 172% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 265% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 99% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 171% | 0% |
Decisions made in chat evaporate — three weeks later no one remembers what was agreed or who owns it. In Claude Cowork this skill reads the real thread, pins down the decision and its owners, and writes a durable record into Notion, with the source linked, so the agreement survives the scroll.
Ask for these if not provided:
Guardrails: don't invent a decision, an owner, or a date not present in the thread; distinguish "decided" from "leaning"; write only to the specified database; if a connector is unauthorised, produce the record inline and say what couldn't be written.
A Decision Record:
> the choice made — one line
Decided by: … · Date: … · Reversibility: one-way / two-way
| Owner | Action | By | |---|---|---|
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,061 | 14,044 | +177% | 1 | 1 | 0% | 802 | 2,108 | +163% | 0 | 0 | — |
case-02 | fail→fail | 11,276 | 19,494 | +73% | 1 | 1 | 0% | 1,107 | 2,854 | +158% | 0 | 0 | — |
case-03 | fail→fail | 13,348 | 16,419 | +23% | 1 | 1 | 0% | 461 | 2,346 | +409% | 0 | 0 | — |
case-04 | fail→fail | 8,336 | 10,787 | +29% | 1 | 1 | 0% | 1,127 | 2,261 | +101% | 0 | 0 | — |
case-05 | fail→fail | 12,191 | 14,193 | +16% | 1 | 1 | 0% | 1,224 | 2,260 | +85% | 0 | 0 | — |
case-06 | fail→pass | 10,402 | 12,327 | +19% | 1 | 1 | 0% | 703 | 1,912 | +172% | 0 | 0 | — |
case-07 | fail→pass | 12,656 | 13,546 | +7% | 1 | 1 | 0% | 1,649 | 2,069 | +25% | 0 | 0 | — |
case-08 | fail→pass | 18,881 | 11,413 | -40% | 1 | 1 | 0% | 481 | 1,755 | +265% | 0 | 0 | — |
case-09 | fail→fail | 17,730 | 18,232 | +3% | 1 | 1 | 0% | 1,652 | 2,949 | +79% | 0 | 0 | — |
case-10 | fail→pass | 11,746 | 11,782 | +0% | 1 | 1 | 0% | 999 | 1,990 | +99% | 0 | 0 | — |
case-11 | fail→pass | 9,413 | 10,319 | +10% | 1 | 1 | 0% | 694 | 1,883 | +171% | 0 | 0 | — |
case-12 | fail→pass | 17,088 | 15,907 | -7% | 1 | 1 | 0% | 1,219 | 2,310 | +89% | 0 | 0 | — |
case-13 | pass→fail | 10,439 | 11,762 | +13% | 1 | 1 | 0% | 911 | 2,123 | +133% | 0 | 0 | — |
case-14 | fail→pass | 14,640 | 15,698 | +7% | 1 | 1 | 0% | 1,485 | 2,606 | +75% | 0 | 0 | — |
case-15 | pass→pass | 14,910 | 13,898 | -7% | 1 | 1 | 0% | 1,350 | 2,157 | +60% | 0 | 0 | — |
case-16 | pass→pass | 19,425 | 14,718 | -24% | 1 | 1 | 0% | 3,088 | 2,205 | -29% | 0 | 0 | — |
case-17 | pass→pass | 11,756 | 34,437 | +193% | 1 | 1 | 0% | 1,008 | 4,511 | +348% | 0 | 0 | — |
case-18 | fail→fail | 14,429 | 15,795 | +9% | 1 | 1 | 0% | 1,219 | 2,622 | +115% | 0 | 0 | — |
case-19 | fail→pass | 35,351 | 11,361 | -68% | 1 | 1 | 0% | 540 | 1,862 | +245% | 0 | 0 | — |
case-20 | fail→fail | 7,446 | 22,697 | +205% | 1 | 1 | 0% | 342 | 3,420 | +900% | 0 | 0 | — |
case-21 | pass→pass | 22,099 | 20,476 | -7% | 1 | 1 | 0% | 2,653 | 3,340 | +26% | 0 | 0 | — |
case-22 | pass→pass | 16,557 | 22,139 | +34% | 1 | 1 | 0% | 1,637 | 2,256 | +38% | 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 20 counted toward the lift figure. The other 2 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 +32 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.