Skip to content
Back to skills

Map Independent Acceptance Evidence

ASecurity

Map completion claims to evidence with separate provenance. Use when several reports or checks may all depend on the same producer, fixture, or flawed source.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 26, 2026
ai-agents

Security analysis

A100/100

Scanned September 26, 2026

npx -y skills add alisonjieli-png/loop-engine --skill map-independent-acceptance-evidence --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Map Independent Acceptance Evidence?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Map Independent Acceptance Evidence
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/alisonjieli-png-map-independent-acceptance-evidence/badge)](https://www.skillsdirectory.com/skills/alisonjieli-png-map-independent-acceptance-evidence)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: map-independent-acceptance-evidence
description: Map completion claims to evidence with separate provenance. Use when several reports or checks may all depend on the same producer, fixture, or flawed source.
---

# Map independent acceptance evidence

## Use

Use before accepting an agent-produced result when many apparent checks may share one source of error. Work from the supplied claims and evidence records. This skill prepares a verification plan; it does not award approval.

## Inputs

- Acceptance conditions and the produced artifact references.
- The claimed checks, their runners, inputs, fixtures, and output records.
- Known author or producer identities and any evaluator independence requirement.

## Procedure

1. Give every acceptance condition a separate row. Record the smallest observable fact that would satisfy it.
2. Trace each proposed proof back through its producer, fixture, checker, and source data. Two reports from the same run count as one evidence origin.
3. Mark circular proofs, such as a generated answer graded by the generator's own assertion, and shared failure points, such as two checkers reading one stale fixture.
4. For each condition, identify an available independent artifact or a specific missing check that could falsify the claim.
5. Keep artifact presence, successful execution, and accepted behavior separate. A clean exit or a producer summary is never a substitute for the required artifact or outcome.
6. Return an evidence map with condition, claimed proof, provenance chain, independence finding, and next verification action.

## Completion check

Every condition has an independent evidence path or an explicit gap. Shared evidence origins and conflicts are visible. The map distinguishes proposed verification from verification actually performed.

## Stop

If the evidence provenance cannot be established, mark the claim unverified. Do not label agreement among dependent reports as independent confirmation.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.