Operating index
verify · operating skill
Evidence-first Review
Turn a claimed result into explicit checks, inspect the evidence, and separate verified facts from unresolved items before approval.
The copied instruction points your AI to a self-contained public Skill Bundle. You can stay on this human-readable page; opening raw source is optional.
When to reach for it
- An agent says an implementation, analysis, migration, or release task is complete.
- The result has observable acceptance criteria or artifacts that can be checked.
- A plausible narrative is not enough to justify approval.
Do not use when
- The task is purely exploratory and has no agreed success condition yet.
- The required source, system, or verification tool is unavailable; record the limitation instead of pretending to verify.
Inputs
- Claimed result or completion report
- Acceptance criteria or intended outcome
- Available evidence such as source files, tests, screenshots, logs, URLs, or system state
Outputs
- Verified claims
- Unverified or failed claims
- Evidence references
- Residual risk and recommended next action
What it does
Evidence-first Review changes the review question from “does this sound finished?” to “what can I actually verify?” It is designed for AI-assisted work where confident completion language can arrive before the underlying state has been checked.
When to reach for it
Use it at a verification or approval boundary: after implementation, after a migration, before a release, or whenever an agent reports that a task is done.
What one run does
- Extract the material claims from the completion report.
- Map each claim to an observable check.
- Inspect the available evidence.
- Mark each claim as verified, failed, or unresolved.
- Report residual risk and the smallest next action.
Common failure modes
- Treating a successful command as proof of the entire user-visible outcome.
- Repeating an agent’s own completion summary as evidence.
- Hiding unavailable checks inside a generally positive verdict.
- Expanding scope during review instead of separating follow-up work.
It’s working if
A reviewer can see exactly why a claim is accepted, what remains uncertain, and what should happen next without reconstructing the whole session.
Where it fits
This is a Verify-phase skill. It pairs naturally with a scoped implementation workflow. A public-safe publishing skill may become a downstream step once that capability is actually published.
Verification
- Every approved claim points to observable evidence or an explicit check result.
- Failed or unavailable checks remain visible in the final review.
- The next action is clear without rereading the full work log.
Guardrails
- Never upgrade an unobserved claim to verified.
- Distinguish source evidence from inference.
- Stop and report when required verification access is unavailable.