Evidence: cold-agent trial, stock Spec Kit + Clewseau bundle delivers AC-HOME-15 end to end

[rdryfoos]

What this is

A fresh agent with no HomesFlow context was given a working tree with stock Spec Kit re-initialized and the Clewseau bundle installed from the catalog. It was asked to pick up one plain-prose item from the PRD (home display name normalization) and carry it to done. This branch is what it produced, unedited.

Draft on purpose. This PR is evidence for review, not for merge: the trial tree re-initialized .specify/ from stock, so merging would overwrite main's native setup. Review it; don't land it.

Where to look

Commit 1caa3aa carries the trial. The cold agent's own slice is 8 files; everything else in that commit is harness setup (stock re-init plus bundle install) done before the agent started.

The agent's slice	What it shows
HomesFlow.prd.md	Minted FR-HOME-04 and AC-HOME-15 from plain prose: next free numbers, correct grammar, registry index table updated in the same edit
specs/002-home-name-normalize/spec.md	Scope bounded, edge cases enumerated, IDs declared as inherited from the registry, risk table names the silent-gap failure mode
specs/002-home-name-normalize/tasks.md	Every task declares **Traces**:; the proof task is separate from the implementation task; running Gate 2 is itself a task
ios/HomesFlow/Core/Home/HomeDisplayNameNormalizer.swift	Pure helper with @covers FR-HOME-04, AC-HOME-15
ios/HomesFlowTests/HomeDisplayNameNormalizerTests.swift	test_AC_HOME_15_* named to the proof grammar; five assertions covering every edge case the spec lists
docs/clewseau-break-fix-peer-review.md	The break/fix probe: renaming the proof flipped the row to tracked-debt via an open Traces: task, restoring flipped it back

How this differs from a stock Spec Kit PR

Stock Spec Kit produces the spec/plan/tasks artifacts and the code. It does not produce: durable IDs minted into a registry with an updated index, Traces: on every task, @covers annotations in source, a proof named to a checkable grammar, or a machine-checkable emit (clew.json) proving the chain held. Every one of those is present here, produced by an agent that had never seen this repo, guided only by what the bundle installed.

Honest caveats

• Gate 2 verifies that a named proof exists, not that it passes. The test's assertions were executed independently during review (5 of 5 pass), but this repo has no CI, so proof execution is attested by review, not by pipeline.
• Two "if needed" tasks (T004, T006) are checked without evidence they ran. The Gate audits AC proofs and open debt carriers, not the honesty of closed checkboxes.
• The agent chose not to mint a US for a pure helper with no user-facing behavior, cross-referencing FR/AC from the stories section instead. Reasonable altitude call; noting it because the trial commit message says "minted US/FR/AC", which overstates.

Made with Cursor

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 12384ad0-d689-476e-8078-f283c7ed120c

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}