agentsclimarketplace

Find the keystone

Skill zakelfassi/skdd-commons/packs/2026-07-growth-circle/find-the-keystone

Skills that evolve in public — community agent-skill drops for SkDD

Install
npx -y skills add zakelfassi/skdd-commons --skill find-the-keystone

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

  • 0 stars0 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.

What its author says it does

Copied from the file, not written here

Derive a product's keystone metric — who × does-what × N times × window — by working backwards from engaged users, then stress-test it for causation and gameability before anyone reorganizes around it. Use when a team asks "what's our North Star", when activation work has no target, when an existing keystone was adopted from a blog post rather than cohort evidence, or when GROWTH.md has an Aha but an empty Keystone.

SKILL.md

4.0 KB, 785 tokens by cl100k_base, as published. Nobody here has run it

Find the Keystone

Work backwards from your most engaged users to one metric that operationalizes core value — then try to break it before you build on it.

Gate

Requires gate: aha = open in GROWTH.md (non-empty ## Aha, not NONE FOUND). If closed, stop and run name-the-aha first — a keystone without a named aha is a number in search of a meaning.

Inputs

  • GROWTH.md (## Aha), and event/cohort data: analytics export, SQL access, or server logs. No data → stop, route to instrument-the-truth for a minimal keystone-path event trail, resume after first light.

Steps

  1. Define "engaged" from the aha, not from activity. Pick the behavior that IS repeated core value delivery (weekly voice note sent, not weekly app open). Select a broad cross-section of users who sustain it.

  2. Path backwards. For each engaged user, reconstruct the first N days: what did they do, in what order, how fast? Contrast with a matched cohort that signed up and died. You're hunting the earliest common fork.

  3. Compress to the four fields. who × does-what × how-many-times × within-window. If it needs a fifth clause, it's a dashboard, not a keystone. Prefer legible-to-everyone over statistically maximal.

  4. Try to kill it — correlation first. The classic critique: the early behavior may be the symptom of intent, not the cause of retention. Design the cheapest causal probe — an experiment that pushes marginal users over the threshold and watches whether retention follows. Pre-register it in the experiment ledger; set causal-probe status.

  5. Try to kill it — gameability second. Red-team: how would a well-meaning team inflate this number while destroying value? (Auto-add connections. Prompt pressure.) Write the top 2 gaming vectors next to the metric — they are the anti-Goodhart tripwires.

  6. Write the spec with a sunset. To ## Keystone: the four fields, derivation evidence, causal-probe status, gaming vectors, and a review date (default: 2 quarters). Set gate: keystone = open. A keystone without planned obsolescence is Goodhart fuel.

Conventions

  • One keystone. Contexts may read it differently (reset-for-context), but the product has one center of gravity at a time.
  • Never adopt a keystone by analogy ("we're like Slack, so 2,000 messages"). Analogy proposes; only your cohorts dispose.
  • Legibility is load-bearing: if the whole team can't recite it, it can't do its actual job, which is organizing people.

Edge Cases

  • Too few users to cohort. Say so. Output a provisional keystone tagged hypothesis, restrict run-the-loop to activation-only experiments, and set the review to "at 500 engaged users".
  • The causal probe fails. A success of the method. Log the busted candidate in the lore ledger with its epitaph and return to step 2 for the next common fork.
  • Two candidates tie. Pick the one the team can recite and affect.
  • Sunset date passed. Re-run steps 1–5 against fresh cohorts before any other growth work; a stale keystone silently closes its gate.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most product growth skills give in 785 tokens

Counted across 728 of the 1,010 authors here whose files we hold, read 2026-08-07

  • Read product marketing context before asking questionsin 24 of 728, across 18 files
  • Define the ideal customer profilein 21 of 728, across 3 files
  • Document a rollback plan before deploymentin 21 of 728, across 12 files
  • Analyze the codebase to understand the productin 19 of 728, across 1 file
  • Ask clarifying questions about the value propositionin 19 of 728, across 1 file
  • Search for companies matching the criteriain 19 of 728, across 1 file
  • Look for signals of immediate needin 19 of 728, across 1 file
  • Assign a fit score from one to tenin 19 of 728, across 1 file
  • Identify the target decision-maker rolein 19 of 728, across 1 file
  • Suggest a personalized contact strategyin 19 of 728, across 1 file
  • Provide conversation starters for outreachin 19 of 728, across 1 file
  • Format results in a scannable markdown templatein 19 of 728, across 1 file

Said here and by no other author read

  • require open aha gate
  • route to instrument-the-truth without data
  • define engaged from aha not activity
  • path backwards from engaged users
  • compress metric to four fields
  • pre-register causal probe in experiment ledger

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.