verify-inherited-premise
Skill3 starsChanged 3 months agoMITView on GitHub
Use when dispatching a task (or about to act) whose premise comes from an inherited claim rather than direct observation of the live system — a handoff/HANDOFF.md doc, an inbox or spec artifact, a skill's own documented schema/paths/behavior, a prior agent's summary, or a PR-body assertion. Adds a fidelity guard: verify the load-bearing claim against reality FIRST, and STOP-and-report if it's false instead of building on it or fabricating to fill the gap. Fires on "verify the premise", "add a fidelity guard", "don't trust the doc", or any dispatch whose success hinges on a claim not yet checked against the actual system.
In this repository41 other files it gives its agents.
- cadence-goalsSkill
- ci-feedback-loopSkill
- compoundSkill
- decision-rights-doctrineSkill
- diataxisSkill
- dispatch-gateSkill
- drift-checkSkill
- family-checkSkill
- fix-then-re-review-ladderSkill
- github-projectsSkill
- gitSkill
- goal-treeSkill
- handoffSkill
- improvement-loopSkill
- kb-captureSkill
- kb-compileSkill
- kb-coreSkill
- kb-lessons-bridgeSkill
- kb-lintSkill
- l1-supervisorSkill
- l2-supervisorSkill
- lessons-learnedSkill
- lN-lifecycle-doctrineSkill
- lN-review-doctrineSkill
- medical-recordsSkill
- operator-interview-doctrineSkill
- operator-interviewSkill
- operator-reviewSkill
- planning-workflowSkill
- pr-review-fanoutSkill
- pull-taskSkill
- ralph-wiggumSkill
- reviewSkill
- root-cause-depthSkill
- scan-documentSkill
- self-improvementSkill
- self-verifySkill
- skillifySkill
- task-workflowSkill
- testing-without-mocksSkill
- vault-cloudSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 31
When verifying yourself, cite reality, not the doc: Prefer "verified against <artifact>" over "per the doc." If a metric is load-bearing …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
verify-inherited-premise
Inherited claims drift from reality. A handoff says "the PR disclosed X"; the PR doesn't. A skill documents a vault schema that doesn't exist. A worker reports "resolved" when it wasn't. A load-bearing metric is a counting artifact. Building on an unverified premise wastes the work — or ships something wrong. This makes premise-verification an explicit, mechanical step of any dispatch.
When to use
Before dispatching a worker, or acting yourself, when success depends on a claim you have NOT checked against the live system. Highest value when the premise came from: a handoff/session-summary doc; an inbox item, spec note, or design artifact; a skill's self-description (its documented templates, schemas, paths, flags); a prior agent's or your own earlier summary; a PR body, comment, or commit message.
Steps
- Name the load-bearing premise(s). Which claim, if false, wastes or misdirects this work? List those, not every assumption.
- Trace each to its source. Mark any sourced from a document/summary rather than direct observation.
- Add a fidelity guard to the dispatch. Instruct the worker to verify each document-sourced premise against the actual artifact FIRST — read the file, the PR body, the live schema, run the command — before building on it. Name the specific check.
- Require STOP-and-report on mismatch. The worker halts with an explicit
STOP — <premise>: <what disagrees>and returns, rather than improvising a fix, paraphrasing over a gap, or proceeding on a wrong assumption. Fabricating to fill a gap is the failure this prevents. - When verifying yourself, cite reality, not the doc. Prefer "verified against
<artifact>" over "per the doc." If a metric is load-bearing for a decision, confirm what it actually measures before concluding.
Edge Cases
- Cheap, directly observable premise — check it inline; no worker guard needed.
- Operator-supplied claim — still verify checkable facts (a spec built on it must be correct), respectfully; don't treat operator decisions/preferences as premises to "verify."
- Load-bearing vs incidental — guard the premise whose falsity breaks the task; don't bloat the dispatch verifying trivia.
- Over-correction — verifying a wrong claim tempts an equally-wrong opposite conclusion; if the check is ambiguous, mark UNCERTAIN and quote the evidence rather than forcing a new confident claim.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

