How you conduct
You conduct like The Smith-Conductor: Verification is your home turf, with 736 agents this window, with a tooling streak underneath.
@rrocha-eng ✓ · Rafael Rocha
Rigor that won't just pass the test — it forges the bench that guarantees the green.
Your blend
One score per tool
Your scores by harness
Each harness is scored only on what it can structurally observe (κ = observable capacity) and ranked against others on the same tool.
The fleet
736
AI agents conducted in the last 30 days
65
multi-agent workflows
2*
peak agents in parallel
36
work sessions
* Peak parallelism is a hardware ceiling — concurrency is capped by the machine's CPU cores, not by skill. The scale of the operation is the fleet total above.
Your working style, read from the evidence
How you conduct
A metacognitive mirror built from your own structural signals — never a personality quiz. deterministic
Your unfair advantage
Your edge is Verification — your own signals show it (verification at 56%, a fleet of 736).
Your next frontier
The most exciting move ahead is Autonomy: it's where you have the most open room, and a single new habit grows fast there.
Mastery zone
Verification
Verification is a strength (14.4/18): the signals back it.
→ You verify 56% of edits → run a test/lint right after editing to reach 80%.
Tooling
Tooling is a strength (12.5/16): the signals back it.
→ 6 MCP servers in play → connect one more to widen your tooling.
Competent
Orchestration
Orchestration (23.0/30) is solid, not your signature.
→ Peak now 2 concurrent agents → dispatch one more agent in the same verified workflow.
Craft & Discipline
Craft (8.9/12) is solid, not your signature.
→ 36 compactions → shorter, focused sessions cut noise and lift craft.
Most room to grow
Delivery
Delivery (9.0/14) is your blind spot — your biggest lever.
→ 15 PRs this window → close the loop by opening a PR when each task lands.
Autonomy
Autonomy (3.5/10) is your blind spot — your biggest lever.
→ Longest hands-off run: 12 steps → delegate a bigger block with automated verification.
What's holding you back
Autonomy is where you have the most open room ahead (afk_max_run=12). One step here next window grows more than reinforcing what's already strong — it's the most generous frontier you have right now.
afk_max_run=12
Where you fit
Orchestrators · quality-owning lead
- high-stakes correctness work
- refactors under risk
Complements Parallel Maestro, The Explorer
Rubs against The Autopilot
Your next mission
Over the next 30d, raise afk_max_run from 12 to 40 and unlock the Autopilot archetype.
The Autopilot
Auditable, unlocked from your signals
Discoveries
Sub-archetype
The Verifier
MCP Explorer
Heavy Fleet
Workflow-dense
Tool Scout
Centurion
Workflow Pilot
Broad Verification
Connected
Toolsmith
Shipper
Self-scheduler
Pillar breakdown
Each pillar score, with the evidence that sustains it.
-
Orchestration 23.0/30
736 agents dispatched across 65 workflows, peaking at 2 running at once.
-
Verification 14.4/18
56% of editing sessions verified, across 4 verification modes, with 27 human sign-offs.
-
Tooling 12.5/16
21 distinct tools in play, 419 advanced tool moves, 6 external systems via MCP.
-
Delivery 9.0/14
15 pull requests and 131 files changed, across 4 codebases.
-
Craft & Discipline 8.9/12
36 context compactions and 20 API errors over 36 sessions.
-
Autonomy 3.5/10
3 self-scheduled loops and a longest hands-off run of 12 agent steps.
Harnesses measured
Which agent harnesses this profile draws on — and what each one's journal can structurally prove. Disclosure only; it never changes the score.
| Harness | Orch | Verify | Tools | Ship | Craft | Auto |
|---|---|---|---|---|---|---|
| Claude Code | full | full | full | full | full | full |
A "partial" or "none" cell means the harness simply doesn't record that signal — the pillar is measured from what it does record, never penalised for what it can't.
Last 30 days
Signature stats
The raw signals behind the score — plain language, no jargon.
736
AI agents conducted
65
Multi-agent workflows
2
Peak agents in parallel
hardware-capped
15
Pull requests shipped
131
Files changed
56%
Edit sessions verified
12
Longest hands-off run
consecutive agent steps
19/30
Active days (of 30)
78%
Evenly spread work
4
Distinct codebases
6
External systems (MCP)
21
Distinct tools used
419
Advanced tool plays
27
Human sign-offs requested
Archetype
The Smith-Conductor
Base form: The Deliberate
The Deliberate leads with craft and verification — nothing ships without proof.
How they work
Every change earns its place: tests, builds and re-reads follow the edits, and human sign-off gates the risky steps. Slow is smooth; smooth is fast.
Ecosystem
Tool ecosystem
21
distinct tools
6
external systems (MCP)
419
advanced tool plays
Beyond the core toolkit, 6 external systems — mail, browser, calendar, cloud — are wired in via MCP and driven mid-workflow, alongside 21 distinct tools and 419 advanced tool plays like tool-search and skills.
The human read
For recruiters
What these signals say about how this person works.
Delegates at production scale
736 AI agents conducted through 65 workflows in 30 days. This is someone who multiplies their output by directing machines — not someone typing faster.
Ships with proof, not hope
56% of editing sessions end in verification, and 27 moments paused for explicit human sign-off. Work arrives checked.
Output, not activity
15 pull requests and 131 files changed across 4 codebases — the fleet lands as concrete, reviewable work.
Trusted autonomy
3 scheduled loops and a longest hands-off run of 12 agent steps — autonomy built on checks, not on faith.