pmf-tracker-cc

pmf-tracker-cc

A Markdown-native PMF / founder-research tracker for Claude Code.

Instead of a SaaS dashboard, your product-market-fit process lives as structured, indexed Markdown inside a repo you own. You drive it conversationally: each day you run /pmf-start, Claude resumes where you left off, interviews you about what happened, and writes it down — honestly. No invented numbers, no inflated metrics, evidence weighted and sourced.

What it tracks

The frame is a founder's customer-discovery loop:

• Hypotheses — one file each, with area · signal · status · evidence_weight · sources and the classic Hypothesis / Evidence / Experiment / Result structure.
• Daily logs — what you did, learned, who you talked to, which hypotheses moved.
• Weekly reviews — Focus / Learned / Decisions / Next, rolled up from the week's logs.
• Overview, thesis, scratchpad, and an evidence/trust ledger — the supporting singular docs.
• A generated INDEX.md — status table, grouping by area, reviews, and a "findings so far" section that only ever lists validated hypotheses.

Everything is plain Markdown with YAML front-matter, so it diffs in git, renders anywhere, and is yours.

Commands

Command	What it does
/pmf-start	Begin a daily session: resume recent state, interview you, write today's daily log.
/pmf-wrap	End-of-day: summarize, update the hypotheses the day touched (status changes require your confirmation), capture decisions.
/pmf-review	Roll up the week's logs into a weekly review and refresh the index.
/pmf-index	Regenerate INDEX.md deterministically from the current files.
/pmf-import	One-time import of an existing pmf-data.json tracker into the Markdown layout.

The honesty contract

This tracker is a working reference, not a pitch. The rules in references/honesty-rules.md are non-negotiable and every command obeys them:

• Never fabricate, invent, or inflate. Use TBD, never a placeholder number.
• Mark provenance and evidence weight: ✅ ≥2 independent sources · ⚠️ partial · 🔵 single-source/n=1 · ❌ refuted.
• Don't launder one quote into a trend. Don't change a hypothesis's status without explicit confirmation.
• Append evidence, never overwrite it. If unsure, ask or leave it out.

Install

/plugin marketplace add dennisonbertram/pmf-tracker-cc
/plugin install pmf-tracker-cc

Then, in any repo where you want to track a venture:

/pmf-start

On first run it offers to start empty (a clean pmf/ scaffold) or to copy the fictional Acme Inbox sample so you can see a populated tracker. When you're ready for your own venture, /pmf-start detects the sample and offers to clear it. It also offers to gitignore your private scratchpad.

Local development / dogfooding

To run it from a local checkout (e.g. while developing):

/plugin marketplace add /absolute/path/to/this/repo
/plugin install pmf-tracker-cc

The tracker root is resolved from the current working directory (git rev-parse --show-toplevel → pmf/), so /pmf-start writes into whatever repo you're in — independent of where the plugin is installed.

Importing an existing tracker

If you already track PMF in an HTML/JSON tool that exports a pmf-data.json:

/pmf-import --data /path/to/pmf-data.json

It runs a dry run, asks you to confirm, imports safely (never overwrites existing files), then verifies the round-trip is verbatim. Fields with no source in the original data (evidence_weight, sources, created/updated) are added with an explicit "auto-inferred on import — verify" marker.

Repository layout

.claude-plugin/      plugin.json + marketplace.json (manifest)
skills/              the five /pmf-* commands (one SKILL.md each)
workflows/           pmf-review.js — the typed, parallel weekly rollup
references/          the single source of truth (format spec, honesty rules, evidence vocab, importer contract)
templates/           literal Markdown skeletons the commands instantiate
examples/            a fictional sample tracker (Acme Inbox) — explore it, then clear it to start your own
scripts/             pmf_import.py (import + --verify), pmf_index.py (deterministic INDEX), pmf_md.py (shared helpers)

The scripts are pure Python stdlib — no network, no pip. references/format-spec.md is canonical: if a command and the spec disagree, the spec wins.

License

MIT © 2026 pmf-tracker-cc authors