Cross Review¶

Reference¶

`run_review` — `attune.roundtable.review`¶

def run_review(
    repo_root,                  # str | Path
    seat="codex",               # any SEAT_RECIPES key
    mode="branch",              # "branch" | "staged"
    base_ref="origin/main",
    board=None,                 # attune.roundtable.Board | None
    invoke_seat=...,            # injectable; defaults to the table's runner
) -> dict: ...

Result keys: ok (always True for a completed run), status (findings / clean / absent / format_noncompliant), seat, thread (board thread id), findings (list of dicts with file, line, severity, claim), reply (raw text), manifest (sent / omitted file lists), target (description), board (posted or skipped (<reason>)).

Module constants¶

Name	Value	Meaning
`DEFAULT_SEAT`	`"codex"`	Reviewer seat when none is passed
`DIFF_CAP_CHARS`	`60_000`	Per-run diff budget for the manifest

The reviewer reply budget is ROLE_REPLY_CHARS["reviewer"] (16,000 chars) from attune.roundtable.compiler.

Supporting functions¶

Function	Does
`resolve_target(repo_root, mode, base_ref)`	Read-only per-file diff resolution; raises `ReviewTargetError` on unresolvable targets
`budget_manifest(per_file, cap_chars)`	Packs files under the budget; returns `sent` / `omitted`
`build_brief(target, manifest)`	Renders the reviewer brief with the diff and manifest
`lint_review(text)`	Returns format problems; empty list means compliant
`parse_findings(text)`	Parses `FINDING:` lines into dicts
`ledger_row(result, disposition)`	Renders the receipts.md ledger row

Entry points¶

Surface	Form
Skill	`/cross-review [staged] [seat=codex\\|antigravity]`
Python	`attune.roundtable.review.run_review`

There is deliberately no CLI command, no MCP tool, and no suggested cadence: invocation is manual (chair-ruled, OPEN-2).