Invalidate the lore
Skill zakelfassi/skdd-commons/packs/2026-07-growth-circle/invalidate-the-lore
Skills that evolve in public — community agent-skill drops for SkDD
npx -y skills add zakelfassi/skdd-commons --skill invalidate-the-loreAssembled 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
Harvest a team's beliefs about why users behave, convert each into a testable claim, and run the cheapest disproving test — building an append-only lore ledger of confirmed/busted/undecidable verdicts. Use when roadmap arguments run on gut feeling, before any big bet built on an untested assumption, on a quarterly hygiene cadence, or when someone says "users churn because…" without a link attached.
SKILL.md
3.7 KB, 688 tokens by cl100k_base, as published. Nobody here has run it
Invalidate the Lore
Turn team beliefs into testable claims, test them cheaply, and log the verdicts where nobody can un-log them.
Gate
Works best with gate: instrumented = open, but may run earlier — some
lore dies to data that already exists (support tickets, store reviews, git
history, server logs). No gate blocks collecting lore; only verdicts
need evidence.
Inputs
- Beliefs, harvested from: the user/founder directly ("tell me five things you know about your users"), docs and READMEs, PR discussions, support threads, marketing copy (every headline is a belief).
- Whatever data exists.
GROWTH.md ## Lore ledgeras working memory.
Steps
-
Harvest without judging. Collect beliefs verbatim, attributed and dated. The confident ones and the load-bearing ones first — a belief that steers the roadmap is worth ten idle opinions.
-
Sharpen each into a claim. Rewrite as something falsifiable: "users churn because onboarding is long" → "churned users abandoned during onboarding at a higher rate than retained users." If a belief can't be sharpened, log it
undecidablewith the reason — that's a verdict too. -
Design the cheapest kill. For each claim, the minimum test: usually a query, sometimes a cohort comparison, occasionally a real experiment (route those to
run-the-loop). Order by load-bearing-ness ÷ cost. Independent claims fan out well — one lane per belief, verified separately. -
Run and rule.
confirmed / busted / undecidable, each with an evidence link. Busted lore gets a one-line epitaph. Never bust by counter-assertion — no evidence, no verdict. -
Ledger it. Append to
GROWTH.md ## Lore ledger(date, author, belief, status, evidence). Surface the two or three verdicts that should change what the team does next — a ledger nobody reads is a diary.
Conventions
- Gut feeling isn't banned; it's promoted — from conclusion to hypothesis. Say this out loud when harvesting; it keeps people honest and unhumiliated.
- The ledger is append-only. Corrections are new entries citing old ones.
- Treat your own analyses as lore too: a conclusion this skill produced last quarter is a belief this quarter.
Edge Cases
- The ledger only ever confirms. Audit the sharpening step — soft claims produce soft confirmations. A healthy ledger busts things.
- Belief held by the person paying for the work. Test it exactly as carefully, deliver the verdict exactly as plainly, with the evidence doing the talking.
- Two beliefs conflict. Perfect — at most one survives. Test the pair together and log both verdicts.
- No data whatsoever. Convert the top beliefs into the instrumentation
wishlist and route to
instrument-the-truth; log everything harvested asuntestedso the debt is visible.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.