s3-B1-1 cover Session 3
Spec, build, run — your own /skill on your machine
Backlog Intelligence Build Clinic — spec, build, run
s03 · Betterworks · 60m · PMs ·
released
s03 → 4 blocks, 20 slides, 11 trainer notes, 5 exercises, 2 reflects, 1 quizzes, 2 asides
Pre-read
- Re-skim the 5-step M1 docket from S02 — `prompts/s02/m1-five-step-chain.md`
- Read `prompts/s03/m1-common-errors.md` — 8 failure modes + the one-line fix for each
- Pace slice CSV downloaded locally (same one as S02)
- (Optional) skim the v1 gdoc — we co-author tonight
- Read `prompts/s03/m1-common-errors.md` — 8 failure modes + the one-line fix for each
- Pace slice CSV downloaded locally (same one as S02)
- (Optional) skim the v1 gdoc — we co-author tonight
Session goal
By end-of-session each PM has:
1. **Agreed** the shared v1 spec with the cohort (skill name, named owner, in the gdoc)
2. **Built** that skill on their own machine
3. **Run** it on tonight's slice — own receipt, own findings page
4. **One extension line** for their real Betterworks backlog
1. **Agreed** the shared v1 spec with the cohort (skill name, named owner, in the gdoc)
2. **Built** that skill on their own machine
3. **Run** it on tonight's slice — own receipt, own findings page
4. **One extension line** for their real Betterworks backlog
Recap S02 · frame the arc · setup checks
14 itemsOpen cold. Acknowledge what S02 landed (the race) and what didn't (the spec). Frame the arc: tonight we agree → build → run. Setup checks fast.
s3-B1-2 list Where S02 landed
Two concepts taught. One artifact built. One thread left open.
- Chaining prompts — one job per step, output feeds the next, checkpoint between
- Pick the harness on purpose — same Claude, five shapes (Claude.ai · Code · Cursor · MCP · API)
- You ran the chain — receipts on screen, ~$1, ~8 min
- Spec wasn't agreed — that's the first 15 tonight
- … 1 more
s3-B1-3 activation Before we start — your S02 reflection
~30 sec. Reflection + rating on your S02 page. Builds the muscle.
s3-B1-4 code Two /m1-cleanup runs (1 hour apart)
col 1
col 2
s3-B1-5 list Today
Session 03 · 60 min · PMs · Build Clinic
- Recap S02 · frame the arc · setup checks
- Discussion → v1 spec (co-author the gdoc)
- Clinic — build the skill, run on the slice
- Real-backlog Q&A · takeaways · S04 close
s3-B1-6 code Pre-flight from your terminal
s3-B1-7 activation Drop a 🟢 in chat as each step is green
Red anywhere? DM PG — don't blocker the room.
Discussion → v1 spec (co-author the gdoc)
13 itemsThe first co-authoring block. PG facilitates the five leading questions; cohort takes stances in chat then types in the gdoc. By end of block: agreed v1 spec.
s3-B2-1 section Keep Aha! clean — together
One skill we agree on. One artifact we maintain.
s3-B2-2 list Pain points · S02 discussion
Insights around which to spec the v1.
- Backlog packed with ideas going back years · many unresolved
- Inconsistent tagging — some themed, some not · some Feature-linked, some not
- Duplicates added without checking · built-in checker still needs heavy manual review
- CS adds ideas via Salesforce · ARR-per-vote attached · current SF bug drops some ARR
- … 3 more
s3-B2-3 list Five leading questions
Stance per question. No it-depends.
- What stood out — the one finding from your S02 run worth a leadership conversation
- The mess we keep killing — orphans · dupes · stale items · something else?
- Cadence — weekly run · monthly hygiene · per-release sync?
- Owner — rotating PMOTW · dedicated owner · co-maintained?
- … 1 more
s3-B2-4 activation Co-author the v1 spec in the gdoc
Five fields. Five stances. We lock by 25.
s3-B2-5 list The v1 spec
Five fields in the gdoc — drop your stance under each.
- Name — slash command (e.g.
/aha-hygiene) - Inputs — Aha! Ideas export · Features list · anything else?
- Output — the one artifact (digest · dashboard · markdown report)
- Trigger — manual · scheduled · on-demand?
- … 1 more
Clinic — build the skill, run on the slice
10 itemsEach PM customises m1-cleanup with the agreed v1 fields (name, output format, trigger) and runs it on tonight's slice. PG floats in chat.
s3-B3-1 section Let's give it a spin
Same spec. Use AI to tweak the skills from S02.
s3-B3-2 list Skip the CSV — Aha! has an MCP server
Connect your live Aha! workspace to Claude. No export, no upload.
- Endpoint —
https://{your-aha-domain}.aha.io/api/v1/mcp - Auth — Aha! personal access token (same as the REST API)
- Reads live data — your skill points at Ideas, Features, releases directly
- Docs — Aha! remote MCP server
s3-B3-3 code Tweak the skill with AI — don't hand-edit
Open Claude Code in the skill folder. Tell it what to change.
s3-B3-4 activation Build your /skill and run it on the slice
Three edits · one install · one run · drop your receipt when it lands.
s3-B3-5 activation Where are you in the build?
Drop step + confidence (1–5).
Real-backlog Q&A · takeaways · S04 preview · close
8 itemsThe highest-leverage 10 min. Real-backlog Q&A pulls the room forward. Then takeaways + S04 preview + close.
s3-B4-1 list Today's takeaways
- You agreed the spec — one skill · named owner · in the gdoc
- You built it on your machine — your /[agreed-name], your receipt
- Your extension line — captured in chat · feeds your capstone M1 finding
s3-B4-2 recap Next time
S04 — Friday 16:00 IST · Designers
- AI for Design — Claude as DS guardrail
- The Critic refuses off-system colours, names tokens
- 8 designer prompt templates — your workflow library
s3-B4-3 close Thank you
One thing about your /skill or the spec you'd carry into your team — drop it below.
After this session
- **Pair with your pod's designer by Sunday** — they brief from your M1 finding in S06
- **Read your /skill's findings page tonight** — pick one finding to hand over
- **Brief the designer informally by Wednesday** — one paragraph + one screenshot
- **For S04 (Designers; PMs read-along):** `prompts/s04/pace-design-critic-system-prompt.md` is the parallel pattern for design work — skim if curious
- **Failure-mode recovery (trainer-side):** if running 10+ min over by Block 4, cut the extension-capture chat drop and skip the wrap monologue. Artifacts matter; the closing line doesn't.
---
## Open items (for polish pass / PG review)
1. **Rewrite triggered by S02 partial coverage (covered through slide 27).** Original S03 assumed PMs ran the chain on assigned slices and a v1 spec was agreed in S02. S02 covered the race but not the spec → S03 now does both. Dropped: per-pod slice differentiation (everyone on the same Pace combined slice) and 3-PM mid-session share (the build itself is the share). Added: gdoc co-author beat + customise-from-m1-cleanup build path.
2. **Agreed skill name** — needs to land by min 25. Likely candidates from S02 chat: `/aha-hygiene`, `/backlog-clean`, `/m1-routine`. PG seeds one if convergence is slow.
3. **gdoc state** — confirm `https://docs.google.com/document/d/1XAsTWaVucoIxFrnPpSVzHa_DHp5DqalfoOWfbzl4M7s` has the 5 headings pre-staged and edit-access for anyone-with-link.
4. **`m1-cleanup-skill` template** — confirm cohort still has it from S02 staging (gDrive folder). If not, drop the install one-liner in chat at session start.
5. **Activation count** — 5 exercises across 4 blocks (target ≥3 for 60 min per `_activations.md`). Surfaces: chat-checks, gdoc co-author, chat-build, chat-pulse, zoom Q&A. 4+ distinct types. Passes variety floor.
6. **m1-common-errors as pre-read** — verify the doc lands in the gDrive `s03/` subfolder + chat 24 hrs ahead. Pre-empts ~60% of live unblocking.
- **Read your /skill's findings page tonight** — pick one finding to hand over
- **Brief the designer informally by Wednesday** — one paragraph + one screenshot
- **For S04 (Designers; PMs read-along):** `prompts/s04/pace-design-critic-system-prompt.md` is the parallel pattern for design work — skim if curious
- **Failure-mode recovery (trainer-side):** if running 10+ min over by Block 4, cut the extension-capture chat drop and skip the wrap monologue. Artifacts matter; the closing line doesn't.
---
## Open items (for polish pass / PG review)
1. **Rewrite triggered by S02 partial coverage (covered through slide 27).** Original S03 assumed PMs ran the chain on assigned slices and a v1 spec was agreed in S02. S02 covered the race but not the spec → S03 now does both. Dropped: per-pod slice differentiation (everyone on the same Pace combined slice) and 3-PM mid-session share (the build itself is the share). Added: gdoc co-author beat + customise-from-m1-cleanup build path.
2. **Agreed skill name** — needs to land by min 25. Likely candidates from S02 chat: `/aha-hygiene`, `/backlog-clean`, `/m1-routine`. PG seeds one if convergence is slow.
3. **gdoc state** — confirm `https://docs.google.com/document/d/1XAsTWaVucoIxFrnPpSVzHa_DHp5DqalfoOWfbzl4M7s` has the 5 headings pre-staged and edit-access for anyone-with-link.
4. **`m1-cleanup-skill` template** — confirm cohort still has it from S02 staging (gDrive folder). If not, drop the install one-liner in chat at session start.
5. **Activation count** — 5 exercises across 4 blocks (target ≥3 for 60 min per `_activations.md`). Surfaces: chat-checks, gdoc co-author, chat-build, chat-pulse, zoom Q&A. 4+ distinct types. Passes variety floor.
6. **m1-common-errors as pre-read** — verify the doc lands in the gDrive `s03/` subfolder + chat 24 hrs ahead. Pre-empts ~60% of live unblocking.
Add ?token=… to the URL.