← DKraft CI Review Team
HARNESS.md
This file was scanned for secrets, personal information, and confidential business references before publication. Automated scanning assists but does not guarantee that all sensitive content has been identified. Person names and some location references are not automatically detected. You are responsible for reviewing this file before making it public.

DKraft CI Review Team — Harness (v3)

Topology

  • Pattern: swarm + router
  • Agents: 3
    • triage (Claude Haiku) — classifies diff size/risk; skips docs-only + lockfile-only PRs
    • codepilot-cr (Claude Sonnet) — correctness + style lens
    • codepilot-cr-perf (Claude Sonnet) — performance + allocation lens

Flow

  1. Triage classifies the PR; trivial diffs get a one-line ack and skip deep review
  2. Non-trivial diffs go to both lenses in parallel; findings merged + deduped
  3. Post comments; block merge on any "critical" finding

Notes

  • Haiku triage cut per-PR cost by routing ~40% of PRs around the expensive lenses

Version history (3)

How HARNESS.md evolved across versions.

  1. v3Jun 22, 2026illustrative

    Put a cheap Haiku triage step in front: docs-only and lockfile-only PRs get a one-line ack and skip the expensive lenses. Roughly 40% of PRs now route around deep review.

    ▸View this version's HARNESS.md
    # DKraft CI Review Team — Harness (v3)

## Topology
- Pattern: swarm + router
- Agents: 3
  - triage (Claude Haiku) — classifies diff size/risk; skips docs-only + lockfile-only PRs
  - codepilot-cr (Claude Sonnet) — correctness + style lens
  - codepilot-cr-perf (Claude Sonnet) — performance + allocation lens

## Flow
1. Triage classifies the PR; trivial diffs get a one-line ack and skip deep review
2. Non-trivial diffs go to both lenses in parallel; findings merged + deduped
3. Post comments; block merge on any "critical" finding

## Notes
- Haiku triage cut per-PR cost by routing ~40% of PRs around the expensive lenses
  2. v2May 20, 2026illustrative

    Added a second review lens focused on performance and allocations, running concurrently with the correctness lens; findings are merged and deduped.

    ▸View this version's HARNESS.md
    # DKraft CI Review Team — Harness (v2)

## Topology
- Pattern: swarm
- Agents: 2 (run in parallel)
  - codepilot-cr (Claude Sonnet) — correctness + style lens
  - codepilot-cr-perf (Claude Sonnet) — performance + allocation lens

## Flow
1. On PR open, pull the diff
2. Both lenses review concurrently; findings are merged + deduped
3. Post comments; block merge on any "critical" finding

## Notes
- Added a dedicated performance lens after regressions slipped past v1
  3. v1Apr 8, 2026illustrative

    First CI reviewer: one model reviews the whole diff and blocks merge on any critical finding.

    ▸View this version's HARNESS.md
    # DKraft CI Review Team — Harness (v1)

## Topology
- Pattern: swarm
- Agents: 1
  - codepilot-cr (Claude Sonnet) — reviews the full diff, posts inline comments

## Flow
1. On PR open, pull the diff
2. Model reviews every changed hunk
3. Post comments; block merge on any "critical" finding

## Known issues
- Reviews trivial diffs (docs, lockfiles) at full cost
- One lens: performance regressions slip through

Compare versions

Diff HARNESS.md between any two captured versions — not just the latest change.

Comparevsillustrative
v1 → v3+10 −9
- # DKraft CI Review Team — Harness (v1)+ # DKraft CI Review Team — Harness (v3)     ## Topology- - Pattern: swarm- - Agents: 1-   - codepilot-cr (Claude Sonnet) — reviews the full diff, posts inline comments+ - Pattern: swarm + router+ - Agents: 3+   - triage (Claude Haiku) — classifies diff size/risk; skips docs-only + lockfile-only PRs+   - codepilot-cr (Claude Sonnet) — correctness + style lens+   - codepilot-cr-perf (Claude Sonnet) — performance + allocation lens     ## Flow- 1. On PR open, pull the diff- 2. Model reviews every changed hunk+ 1. Triage classifies the PR; trivial diffs get a one-line ack and skip deep review+ 2. Non-trivial diffs go to both lenses in parallel; findings merged + deduped  3. Post comments; block merge on any "critical" finding   - ## Known issues- - Reviews trivial diffs (docs, lockfiles) at full cost- - One lens: performance regressions slip through+ ## Notes+ - Haiku triage cut per-PR cost by routing ~40% of PRs around the expensive lenses