🔬

DKraft CI Review Team

Self-ReportedIllustrative

Three-lens parallel PR review: security, performance, test coverage.

Dana Kraft· Operating since Dec 1, 2025· active
Follow1 follower
This team is illustrative — clearly labeled demo content, not a claim to a real deployment. Used to demonstrate breadth of industry coverage. Never counted in the real or curated totals.

Recent activity

Version cuts and proof, newest first — the living track record.

  1. Version · Cut v3 — v3 — Haiku triage routerillustrative2mo ago
  2. Milestone · Shipped v3 — Haiku triage routerillustrative2mo ago
  3. Milestone · 1,000 PRs reviewed with zero missed HIGH-severity findingsillustrative3mo ago
  4. Version · Cut v2 — v2 — parallel performance lensillustrative3mo ago
  5. Lesson · One review lens misses performance regressionsillustrative3mo ago

Spec sheet

The benchmark fields — designed for comparison across teams.

Topology
Swarm
Agent count
3
Platform
LangGraph
Runs on
LangGraph ×2 · Custom
Industries
software-delivery
Task kinds
code-reviewsecurity-auditperformance-analysistest-coverage
Trust tier
Self-Reported
Proof entries
0

Topology & roster

Swarm

Peer parallel review with synthesis. Three CodePilot CR instances each run a distinct lens (security, performance, test coverage) simultaneously; LedgerLine tracks findings and cost allocation; a synthesis step merges findings. (Fictional.)

System wiring

Typical Swarm layout — schematic, not verified wiring
Typical role-level schematic — not verified wiringdirectsdirectsdirectscoordinates viacoordinates viacoordinates viaHuman operatorHumanoperatorHUMANGATEPeer worker APeer worker ABUILDERPeer worker BPeer worker BBUILDERPeer worker CPeer worker CBUILDERMessage busMessage busMESSAGE BUS
Node details

Typical Swarm layout — schematic, not verified wiring

HumanHuman operatorHuman gate
Tool
Human operator
Autonomy
Human-gated
Sends
  • directs → Peer worker A
  • directs → Peer worker B
  • directs → Peer worker C
BuilderPeer worker A
Tool
Peer worker A
Autonomy
Runs autonomously
Sends
  • coordinates via → Message bus
Receives
  • directs ← Human operator
BuilderPeer worker B
Tool
Peer worker B
Autonomy
Runs autonomously
Sends
  • coordinates via → Message bus
Receives
  • directs ← Human operator
BuilderPeer worker C
Tool
Peer worker C
Autonomy
Runs autonomously
Sends
  • coordinates via → Message bus
Receives
  • directs ← Human operator
Message busMessage bus
Tool
Message bus
Autonomy
Runs autonomously
Receives
  • coordinates via ← Peer worker A
  • coordinates via ← Peer worker B
  • coordinates via ← Peer worker C

How a typical Swarm team handles a task

Typical Swarm layout — schematic, not verified wiring

  1. Task arrives

    Human operator directs Peer worker A, Peer worker B, and Peer worker C.

  2. The builders execute

    Peer worker A, Peer worker B, and Peer worker C build the work. Coordination flows over Message bus.

  3. Human holds the last word

    Human operator holds final approval.

Replicate a typical Swarm setup

Typical Swarm layout — schematic, not verified wiring

Ingredients

  • HumanHuman operator
  • BuilderPeer worker A
  • BuilderPeer worker B
  • BuilderPeer worker C
  • Message busMessage bus

Setup order

  1. 1.Provision the substrate: Message bus.
  2. 2.Wire Peer worker A: it receives "directs" from Human operator. Wire Peer worker B: it receives "directs" from Human operator. Wire Peer worker C: it receives "directs" from Human operator.
  3. 3.Declare the human gate: Human operator holds final approval.
🚦
CodePilot CRSecurity Reviewerillustrative

LangGraph · 0.2.74

Claude Sonnet 4.6 · 2026-02

⚡
CodePilot CR (Performance)Performance Reviewerillustrative

LangGraph · 0.2.74

Claude Sonnet 4.6 · 2026-02

🧾
LedgerLineFindings Trackerillustrative

Custom · internal 2.x

GPT-4o mini · 2024-07

Performance metrics

Windowed metrics with provenance. [unknown] means it was not tracked — an honest hole beats an invented figure.

PRs reviewed
1,240
self-reportedillustrative
as of Jun 1, 2026
Finding acceptance rate
89.4%
self-reportedillustrative
as of Jun 1, 2026

Token economics

Cost transparency is part of the honesty architecture. [unknown] means it was not tracked — not that it is zero.

No cost metrics on record. Cost tracking is hard across runtimes; honest absence beats invented figures.

Blueprint

Operational DNA — why it works, how it was built, and how it is overseen. Not files for sale; knowledge of the design.

Why it works

Parallel lens separation prevents a single reviewer from gravitating to one issue type. Each reviewer's scope is narrow enough to achieve depth without coverage gaps. Synthesis deduplication prevents the same finding being reported three times. (Fictional.)

How it was built

LangGraph fan-out: PR diff dispatched to three parallel reviewers; findings collected into structured JSON; synthesis model merges and deduplicates. (Fictional.)

Oversight model

Findings above severity HIGH require human sign-off before merge. (Fictional.)

Version history (3)

Each version is a snapshot of the public harness files plus the changelog note — the operational DNA over time, with the metric that moved.

  1. v3v3 — Haiku triage routerJun 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.

    cost per PR$0.42→$0.19
    ▸Diff vs v2 — +7 −6 across 1 file
    HARNESS.md+7 −6
    - # DKraft CI Review Team — Harness (v2)+ # DKraft CI Review Team — Harness (v3)     ## Topology- - Pattern: swarm- - Agents: 2 (run in parallel)+ - 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. Both lenses review concurrently; findings are merged + deduped+ 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- - Added a dedicated performance lens after regressions slipped past v1+ - Haiku triage cut per-PR cost by routing ~40% of PRs around the expensive lenses
  2. v2v2 — parallel performance lensMay 20, 2026illustrative

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

    perf false-negative rate18%→7%
    ▸Diff vs v1 — +7 −7 across 1 file
    HARNESS.md+7 −7
    - # DKraft CI Review Team — Harness (v1)+ # DKraft CI Review Team — Harness (v2)     ## Topology  - Pattern: swarm- - Agents: 1-   - codepilot-cr (Claude Sonnet) — reviews the full diff, posts inline comments+ - 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. Model reviews every changed hunk+ 2. Both lenses review concurrently; findings are 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+ - Added a dedicated performance lens after regressions slipped past v1
  3. v1v1 — single-lens reviewApr 8, 2026illustrative

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

    ▸Initial snapshot — +15 −0 across 1 file
    HARNESS.mdnew file
    + # 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

Proof (3)

The team's shared track record — tasks, incidents, lessons, milestones. Per-entry provenance tags are always visible.

  1. MilestoneJun 22, 2026evidence-linkedillustrative

    Shipped v3 — Haiku triage router

    A cheap triage step now skips docs-only and lockfile-only PRs, routing ~40% of pull requests around the expensive review lenses. Per-PR cost fell from $0.42 to $0.19. (Illustrative demo.)

    https://example.com/dkraft/ci-v3
  2. MilestoneMay 25, 2026evidence-linkedillustrative

    1,000 PRs reviewed with zero missed HIGH-severity findings

    https://example.com/dkraft/ci-1000
  3. LessonMay 20, 2026self-reportedillustrative

    One review lens misses performance regressions

    The single-lens v1 kept passing diffs that added O(n²) work in hot paths. Splitting a dedicated performance lens out to run in parallel dropped the perf false-negative rate from 18% to 7%. (Illustrative demo.)

Sign in to add a proof entry.

Sign in

Attestations (0)

Named third-party statements from people with first-hand experience. Attestations are what separates Peer-Attested from Evidence-Linked.

No attestations yet. Worked with this configuration or agent? Attest to it using the form below — attestations are named third-party statements and are what separates Peer-Attested from Evidence-Linked.

Sign in to attest to this team.

Sign in

Files (1)

Operational documents — runbooks, soul files, topology references.