# Completion Proof

> Local report-only detector for completion claims without evidence markers (A11). Does not authenticate evidence or block builds.

- Skill: `zedarvates/completion-proof` (Agent Skill, multi-file: 11 files)
- Install (CLI): `npx get-skill add zedarvates/completion-proof`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zedarvates/completion-proof/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Docs & Writing
- Author: zedarvates (https://skillmd.com/u/zedarvates)
- Updated: 2026-11-11
- Page: https://skillmd.com/skills/zedarvates/completion-proof

---


# Completion proof markers (A11)

Detects reports claiming completion without an associated evidence marker:
`test_id`, `cmd_output_ref`, `artifact_hash`, a passed validation with a
reference, or supported lexical markers in text reports.

```bash
python -m skills.completion_proof.cli path/to/report.json --json
python -m skills.completion_proof.test_completion_proof
```

This is a **report-only heuristic**. A non-empty reference may point to a
nonexistent file; lexical markers can also suppress a finding. The CLI returns
zero even when findings exist. Consumers must inspect `errors` and `findings`.
No test execution, hash verification, or ingestion enforcement is performed.

A separate opt-in `--verify` path now verifies receipt, log and source hashes
against a trusted digest supplied outside the report. See the
[v1 verification contract](VERIFICATION.md). This does not change the legacy
marker audit. Add `--strict` to return nonzero unless verification succeeds;
the calling workflow must honor that exit status before closing its task.

The five finding fields describe evidence, estimated cost, expected benefit,
risk, and a minimal suggested remedy. These are qualitative, not measurements.

See the [replayable demonstration](../../docs/demos/completion-proof/README.md)
and its negative control before treating a clean report as verified success.

