AGM Dogfooding Report

Date: 2026-06-24

Version: v0.1.0

Status: Community Draft

Contributor-side Agent Pass

The contributor-side agent first read the AGM discovery pointer:

  • AGENTS.md

It then read the canonical AGM files:

  • .agm/manifest.yml
  • .agm/risk_zones.yml
  • .agm/evidence_requirements.yml

Demo Change

The selected demo change was a small critical-risk authentication update:

  • demo_app/auth.py: normalize token whitespace before lookup.
  • demo_app/tests/test_auth.py: add a regression test for accidental leading/trailing whitespace.

The changed auth file is classified as critical. The test file is classified as high/configuration.

Tests Actually Run

Command:

python -m pytest demo_app/tests/test_auth.py

Observed result:

4 passed in 0.27s

The test artifact was recorded at:

  • review_packets/artifacts/dogfood_auth_pytest.txt

Evidence Package

The contributor-side simulation generated:

  • evidence_packages/dogfood_auth_pending.yml

The evidence package records:

  • changed files;
  • detected critical/high risk zones;
  • required evidence from AGM;
  • test command and artifact;
  • known limitations;
  • human_review_declaration.status: pending_human_review.

The declaration remains pending because no real human review was performed during this agent-run dogfooding pass. The agent did not fabricate human_reviewed.

Maintainer-side Agent Pass

The maintainer-side agent used the current repository .agm/ files as reference values:

  • base/current .agm/manifest.yml;
  • referenced risk zones;
  • referenced evidence requirements.

It read the submitted evidence package:

  • evidence_packages/dogfood_auth_pending.yml

It generated the diagnostic report at:

  • review_packets/dogfood/review_packet.md
  • review_packets/dogfood/review_packet.json

Reference-vs-observed Findings

The diagnostic report included these governance indicators:

  • AGM reference source: Pass.
  • Governance entrypoint change: Pass, none detected.
  • Risk-zone classification: Pass, authentication and configuration detected.
  • Required tests: Pass, command and artifact provided.
  • Human review declaration: Blocked, observed pending_human_review for critical risk.
  • Placeholder evidence: Pass, absent.
  • Final decision authority: Pass, human decision required.

Gate state:

  • governance_gate_status: blocked
  • technical_review_readiness: not_ready
  • final_acceptance_readiness: blocked_by_policy

This is the expected result for critical risk with pending human review.

Usability Gaps Found

The initial reviewer attention point for incomplete evidence was too artifact/test-specific. In this dogfooding case, the incomplete condition was pending human review. The message was revised to say that human review remains pending and the contribution must not be treated as final-acceptance ready.

Small Corrections Made

  • Updated reviewer attention point generation for pending human review.
  • Regenerated the dogfooding review packet after the correction.

Final decision remains with human maintainers. This dogfooding pass did not approve, reject, or merge the demo change.