Audit receipts for AI coding-agent work

Make agent runs say what evidence is incomplete.

I turn repeated agent workflows into repo-owned instructions plus a small receipt: what was checked, what was skipped, and which claims are only partial or unobservable. Useful for PR review, MCP/tool-call audit logs, and safety checks where pass/fail is not enough.

Get a free audit fit check Read the audit receipt note Open JSON template DM on X

Fit check before work

Good fit if

You already have one repeated agent workflow, and the risky part is whether the final answer proves what was checked, skipped, redacted, or unobservable.

Not for you if

You need a full app build, growth automation, scraping/account work, compliance certification, or a broad prompt pack with no verification boundary.

Example receipt

This is the kind of small record a reviewer should get after the agent finishes.

{
  "workflow": "pr-review",
  "subject": "PR #42",
  "inputs_reviewed": ["diff", "changed files", "test output"],
  "checks_run": ["unit tests"],
  "checks_skipped": ["e2e tests"],
  "evidence_gaps": ["no browser trace", "coverage report missing"],
  "redaction_state": "not_needed",
  "verdict": "incomplete"
}

What you get back

Repo-owned instructions

A small folder with AGENTS.md, a task skill, optional templates, and clear boundaries.

One audit target

PR review, MCP tool calls, safe edits, release notes, or another workflow where evidence matters.

A receipt that can fail honestly

The agent has to say what it checked, what it skipped, and when the evidence is incomplete.

repo/
  AGENTS.md
  skills/
    workflow-name/
      SKILL.md
      verification.md
      templates/
        output-template.md
        completion-receipt.md

What to send

The audit surface

Name the run, tool call, PR review, or safety check where evidence can be incomplete.

The evidence

List what the agent should prove: files reviewed, tool calls observed, redaction state, or checks run.

The gaps

Say what should be allowed to show as partial, unobservable, skipped, or incomplete.

Send an audit fit check Or DM @ChenFN01

What the free fit check returns

Fit / not-fit

A direct answer on whether an audit receipt helps, or whether a normal checklist is enough.

Receipt fields

The smallest fields I would capture: subject, evidence, skipped checks, coverage gaps, and verdict.

Paid pilot judgment

Whether the problem is worth a $99 workflow pack, a $299 setup pass, or no paid work at all.

Before / after

Before

A confident answer with thin evidence.

No issues found.
Audit log exists.
Sensitive values are redacted.
Ready to ship.

After

A receipt that separates outcome from evidence quality.

outcome: no_blocking_findings
files_reviewed: 2
tests_run: none
redaction_state: applied
audit_coverage: partial
verdict: incomplete

Result

A reviewer can see what was checked, what was missing, and whether the gap matters.

Try the audit receipt prompt

Examples

PR review workflow

Find bugs first, cite file lines, avoid style-only comments, and end with an auditable receipt.

Try an audit receipt

Safe edit workflow

Preflight restricted paths, ask before touching sensitive files, and verify what changed.

Open example

Safe receipt workflow

Attach a policy file and receipt schema for checked, blocked, and unverified actions.

Open safety example

Audit receipt JSON

A small machine-readable artifact for CI, MCP audits, and agent runs that may be incomplete.

Open JSON template

MCP team audit receipt

Split local stdio from hosted HTTP identity, shared-token risk, allowlist decisions, and audit coverage gaps.

Open MCP team template

Why audit receipts

A short technical note on why agent runs need `incomplete` verdicts, not longer prompts.

Read article

Your audit workflow

Send one audit surface. I will tell you whether a receipt would help, or if a normal checklist is enough.

Submit audit fit check

Fit-check response example

See the exact kind of fit / not-fit response you get before paying for a pilot.

Open example response

SKILL.md checker

Paste frontmatter and catch common Codex compatibility issues before installing a skill.

Open checker

Public signals so far

These are technical validation threads, not customer logos. I use them to keep the offer grounded.

MCP audit receipt

A ReviewMyMCP maintainer asked to iterate on a machine-readable audit receipt shape.

Open thread

Passive trace coverage

Traceary adopted the split between sensitive match, access operation, redaction state, and coverage gaps.

Open thread

PR contract reports

Forge discussion helped sharpen the run-level contract fields for agent PR review.

Open thread

Pilot

Free fit check

$0

I tell you whether a receipt would help, or whether a normal prompt/checklist is enough.

One audit workflow pack

$99

One audit surface turned into reusable instructions, verification notes, and a receipt template.

Setup + revision

$299

One audit workflow pack, setup guidance, and one revision after you test it.

Paid requests still get a fit/scope reply before any invoice or payment link.

Start with a free audit fit check Request scoped invoice See paid pilot scope DM on X See an example submission