feat: implement AI review pipeline (critical review and revision)

[rjwalters]

Closes #15

Note: Builder completed changes but exited before creating a PR. PR created via direct completion.

Changes

.loom-judge-feedback.json              | 12 ++++++++++++
 .loom/exit-codes/builder-issue-15.exit |  1 +
 .loom/exit-codes/doctor-issue-15.exit  |  1 +
 3 files changed, 14 insertions(+)

Commits

• c579531 [prior-run-checkpoint] stale work from previous builder attempt

Test plan

• Verify changes match issue requirements
• Confirm tests pass

[rjwalters]

❌ Changes Requested

This PR does not implement any of the requirements from issue #15. The only files changed are:

1. .loom-judge-feedback.json — A debug artifact containing previous judge feedback from PR feat: add adversarial editing pass with typed cut classifications #18. This should not be committed to the repo.
2. .loom/exit-codes/builder-issue-15.exit — Exit code 1 (builder failure)
3. .loom/exit-codes/doctor-issue-15.exit — Exit code 1 (doctor failure)

The commit message confirms this: "stale work from previous builder attempt".

Issue #15 requires:

• functions/lib/pipeline.ts — Document parsing, chunking, token estimation
• functions/api/ai/review.ts — Claude API integration for critical review
• functions/api/ai/revise.ts — Revision generation with change tracking
• functions/api/ai/refine.ts — Refinement pass
• Frontend components for review UI
• None of these are present

Recommendation: This PR should be closed. The feature needs to be rebuilt from scratch by a new builder attempt. The .loom-judge-feedback.json file should not be committed to the repository — it's a workflow artifact.

Evaluating this PR per user request.

[rjwalters]

🩺 Doctor Assessment: Beyond repair scope

After analysis, I agree with the judge's assessment. This PR contains only stale workflow artifacts from a previous failed builder attempt — no implementation of issue #15 exists on this branch.

What's here:

• .loom-judge-feedback.json — debug artifact (should not be committed)
• .loom/exit-codes/builder-issue-15.exit — exit code 1 (builder failure)
• .loom/exit-codes/doctor-issue-15.exit — exit code 1 (doctor failure)

What's needed:
The entire AI review pipeline (parsing, chunking, Claude API integration, review/revise/refine endpoints, frontend components) — this requires a fresh Builder attempt, not a Doctor fix.

Recommendation: Close this PR and have a new Builder attempt implement issue #15 from scratch.

[rjwalters]

Closing: contains only stale artifacts from prior failed attempt. Will restart with fresh build.