Audit Protocol

Audit Protocol

Trigger the consensus audit process manually. Evaluates pending evidence and produces verdicts stored in SQLite.

Execute

quorum audit {{ arguments }}

Setup

Read config at {ADAPTER_ROOT}/../../core/config.json:

• audit_submit MCP tool — evidence submission
• consensus.trigger_tag / agree_tag / pending_tag — tag values
• consensus.roles — provider-per-role mapping (advocate, devil, judge)

Options

Flag	Effect
--dry-run	Print the audit prompt without executing
--no-resume	Start a fresh session (discard any saved session)
--auto-fix	After audit, auto-correct rejected items
--model <name>	Override auditor model (default from config)
--reset-session	Delete saved session state before running
--evidence <markdown>	Submit evidence directly (useful for worktree audits)

Verdict Flow

Verdicts are stored in SQLite (not markdown files):

audit → provider reviews evidence
  → verdict stored via bridge.recordTransition()
  → audit-status.json marker written (fast-path for hooks)
  → quorum status shows result

Do NOT look for verdict.md or gpt.md — these files are eliminated. Use quorum status or quorum tool audit_history.

After Completion

quorum tool audit_history --summary

• [agree_tag] — consensus reached, proceed to retrospective then merge
• [pending_tag] — rejection with codes; fix issues, re-submit evidence
• No verdict — audit may have failed; check stderr output

Execution Context

Context	Behavior
Interactive	Run audit, show verdict summary, suggest next steps
Headless	Run audit, output verdict, exit (caller reads audit-status.json)

In headless mode, do NOT ask follow-up questions. Output the result and exit.

Common Rejection Codes

Code	Meaning	Fix
test-gap	Missing or insufficient tests	Add tests covering the claimed changes
claim-drift	Evidence claim doesn't match actual diff	Update claim to match git diff
scope-mismatch	Changed files not listed in evidence	Update Changed Files section
quality-violation	Code quality check failed	Fix lint/type errors