Agent Definitions

Eight Subagents, One Loop

Each agent maps to a phase of the Intent loop. They connect to MCP servers through scoped tool access, use model routing for cost efficiency, and never do each other's work. Knowledge operations shape every decision.


Knowledge Phase

knowledge-compiler

Sonnet
Knowledge — Compile

Compiles raw sources into structured knowledge artifacts. Reads immutable raw/ files, creates or updates personas, journeys, decisions, themes, domain models, and design rationale in knowledge/. Every ingest touches 10–15 artifacts. Maintains the master index, traceability matrix, and activity log.

ingest list_artifacts get_artifact Read Write Bash

knowledge-querier

Haiku
Knowledge — Query & Lint

Queries the compiled knowledge base and runs lint checks. Synthesizes answers with citations to knowledge artifacts and raw sources. Lint detects contradictions, orphans, stale claims, missing cross-refs, and coverage gaps. Each finding becomes a suggested signal for Notice.

query lint list_artifacts get_artifact Read

Notice Phase

signal-capture

Haiku
Notice — Capture

Captures raw signals and computes trust scores. Use when noticing something worth tracking — decisions, risks, requirements, patterns, observations from conversations, code, or agent traces.

Every signal needs clear, specific content (not vague summaries), a source attribution, honest trust factor scoring, and a confidence assessment. A single meeting might produce 3–8 signals of different types.

create_signal list_signals get_signal Read Bash

signal-enricher

Sonnet
Notice — Enrich

Enriches signals: rescores trust, clusters related signals, manages amplification, and promotes clusters to intents. Use when signals need analysis, grouping, or elevation to problems worth solving.

  1. Review unclustered signals (status: captured or active)
  2. Rescore trust if new information warrants it
  3. Cluster signals that share emergent themes
  4. Promote clusters that reveal a problem worth solving
  5. Dismiss signals that are stale, duplicate, or irrelevant
score_trust cluster_signals promote_to_intent add_reference dismiss_signal list_signals get_signal

Spec Phase

spec-writer

Sonnet
Spec — Author

Creates specifications and contracts from intents. Specs are contracts, not stories — precise enough that an AI agent can execute against them autonomously.

Every spec needs a problem statement grounded in signal evidence (cite SIG-NNN IDs), solution description, contracts (4 types), testable acceptance criteria, explicit out-of-scope boundaries, and test scenarios.

create_spec create_contract assess_agent_readiness get_spec list_specs get_signal Read Write Bash

contract-verifier

Sonnet
Execute — Verify

Verifies contracts against implementation. Runs verification commands, inspects outputs, records results. The quality gate between execution and completion.

Critical severity failures block completion. Major severity failures flag for review. Minor severity failures are noted but don't block.

verify_contract get_spec list_specs ingest_event Read Bash

Spec-Shaping Protocol

Intents become agent-ready specs through four-persona interrogation. Each persona queries the knowledge base, checks existing decisions, and generates structured questions and assertions. Brien reviews specs, not execution.

Practitioner-Architect

Pass 1: Shape

Queries DDRs, domain models, rationale. Outputs boundaries, approach, key decisions.

Product-Minded Leader

Pass 2: Outcome

Queries personas, journeys, themes. Outputs why it matters, behavioral change.

Design-Quality Advocate

Pass 3: Contract

Queries DDR validation criteria, existing contracts. Outputs observable outcomes, verification commands.

AI Agent

Pass 4: Readiness

Queries the spec itself + trust formula. Outputs trust score, ambiguity flags, recommended autonomy level.

Trust & Disambiguation If trust < L2 after Pass 4, a disambiguation signal is generated instead of executing. Brien intervenes only when the system can't resolve ambiguity.

Observe Phase

observer

Sonnet
Observe — Close the Loop

Monitors the system, detects deltas, and closes the loop by suggesting new signals from event patterns. The critical feedback mechanism.

Watch for: repeated contract failures, unclustered signal backlogs, specs with no contract verifications, and trust boundary crossings.

ingest_event detect_spec_delta detect_trust_drift system_health suggest_signals_from_events Read Bash

Orchestration

coordinator

Orchestrates the full Intent loop across all agents

The coordinator plans, delegates, and synthesizes. It never does the work itself. It routes to the right agent, in the right order, with the right model. Start every session by asking the observer for system_health to understand the current pipeline state.

Model Routing

Cost efficiency through intelligent model selection. Fast, cheap models for simple capture and queries. Reasoning models for synthesis, judgment, and compilation decisions.

AgentModelRationale
knowledge-compilerSonnetNeeds reasoning for cross-reference synthesis and compilation decisions
knowledge-querierHaikuSimple lookup and pattern matching for queries and lint
signal-captureHaikuCheap, fast — simple capture and trust scoring
signal-enricherSonnetNeeds reasoning for clustering and promotion decisions
spec-writerSonnetNeeds precision for contract definition and completeness
contract-verifierSonnetNeeds judgment for verification result interpretation
observerSonnetNeeds pattern detection for delta and drift analysis
coordinatorOpusOrchestration requires highest reasoning capability
The Voice Library The spec-shaping protocol draws from a library of 178+ thought leader personas — each with their own voice, mental models, and stances. These aren't generic role descriptions; they're calibrated renderings of real thinkers. Meet the 178 voices behind spec-shaping →