Pmf review

Skill dennisonbertram/pmf-tracker-cc/skills/pmf-review

Roll up the week's daily PMF logs into a Weekly Review entry (Focus / Learned / Decisions / Next), surface recommended hypothesis signal/status updates for you to confirm, and refresh the index. Use at the end of a week, or when the user says 'pmf review', 'do my weekly review', 'roll up the week', 'weekly synthesis', or types '/pmf-review'.From its SKILL.md

Install

npx -y skills add dennisonbertram/pmf-tracker-cc --skill pmf-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.7 KB, as published. Nobody here has run it

/pmf-review — weekly rollup

You are synthesizing a week of daily logs into one honest weekly review. Roll up, don't inflate: three n=1 notes on the same point may become n=3, but only say so explicitly and only if justified — never launder repetition into a trend.

Before you do anything

Read ${CLAUDE_PLUGIN_ROOT}/references/honesty-rules.md (NON-NEGOTIABLE), format-spec.md, and evidence-weight.md.
Resolve the tracker root $PMF and the week window: default to the last 7 days ending today (date +%Y-%m-%d); if $ARGUMENTS gives a week-start date, use [start, start+6d].

How to run it

There are two equivalent paths. Prefer the workflow if your environment runs workflow scripts; otherwise do it inline. Both produce the same review.

Path A — workflow (typed, parallel)

The canonical implementation is ${CLAUDE_PLUGIN_ROOT}/workflows/pmf-review.js. It runs four phases — Collect (resolve root + window, list the week's logs, read the prior review's ## Next), Read (one sub-agent per daily log, in parallel, extracting did/learned/convos/movement/decisions with evidence weights preserved — no cross-day synthesis), Synthesize (one agent drafts the four review fields and justifies any evidence-weight rollup), and Write (write the review file, record recommended signal/status changes for you to confirm via /pmf-wrap, regenerate the index). It never edits hypothesis files itself. Invoke it as a workflow if available.

Path B — inline (always works)

Collect. List $PMF/log/*.md whose date is in the window. Read the newest $PMF/reviews/*.md's ## Next.
Read. Read each in-window log. For many logs, fan out one read-only sub-agent per log (Task/Agent tool) to extract {what_did, what_learned, customer_convos, hypothesis_movement, decisions} verbatim with evidence weights — instruct each sub-agent to follow honesty-rules.md and NOT synthesize across days.
Synthesize. From the extracts + the prior ## Next, draft the four fields:
- ## Focus — what the week was actually about.
- ## Learned — grounded findings; roll up evidence weight explicitly (show the n=), no invented trends.
- ## Decisions — decisions actually made.
- ## Next — concrete next steps (carry forward unfinished items from the prior ## Next).
Write. Create $PMF/reviews/<week-start>.md from ${CLAUDE_PLUGIN_ROOT}/templates/weekly-review.md; set period, logs_rolled_up (the filenames), and hypotheses_touched (union across the week). Record any hypothesis signal or status changes the week suggests as recommendations in the review's ## Next (e.g. "Recommend H4 signal — → Mixed; confirm via /pmf-wrap"). Per honesty-rules.md rule 4, neither signal nor status changes to hypothesis files happen here without explicit confirmation — apply them in /pmf-wrap, not in the review. (If you want to apply a change now, switch to /pmf-wrap, which gates it with AskUserQuestion.)

After writing

Regenerate the index: python3 ${CLAUDE_PLUGIN_ROOT}/scripts/pmf_index.py --out "$PMF" --check. Report the review path, which hypotheses changed, and that the index refreshed. If the week had zero logs, say so and write nothing.