Agent Catalog
ASDD defines 10 core specialized agents plus supporting agents for security, observability, and DevOps. Each agent has a precisely defined role, a minimum confidence threshold, and a failure protocol.
Core principle: separation of concerns
Each agent operates within a single, well-defined scope. No agent performs another agent's role. This separation is what makes the pipeline predictable, auditable, and recoverable when an agent fails.
The 10 core agents at a glance
| Agent | Phase | Core Output | Min Confidence | Failure Action |
|---|---|---|---|---|
| Discovery Agent | −1 / 0 | Behavioral slices from raw intent | 0.85 | Draft mode; route to PO/TL |
| Spec Agent | 1 | EARS-format requirements | 0.85 | Flag ambiguous sections; route to TL |
| Validation Agent | 1 | Spec quality score + validation report | 0.90 | Block pipeline; TL sign-off |
| Domain Agent | 2 | Schema-compliant domain model | 0.85 | Draft mode; TL review |
| Design Agent | 3 | Architecture + ADRs + Reasoning Trace | 0.80 | Draft architecture; TL review before proceeding |
| Task Planning Agent | 4 | Execution wave plan (tasks.md) | 0.80 | Draft plan; TL review |
| Implementation Agent | 4 | Code via context-fresh sub-agents | 0.75 | Feature branch; human code review |
| QA Agent | 1 + 5 | Test suites mapped to spec behaviors | 0.85 | Flag uncovered requirements |
| Refactor Agent | 4 | Refactored code within spec boundaries | 0.80 | Flag; TL review |
| Knowledge Agent | All | State manifest, learning proposals | 0.80 | Propose update; human approval |
Supporting agents
| Agent | Role | Min Confidence |
|---|---|---|
| Security Agent | Pre-deployment compliance scan | 0.95 |
| Observability Agent | Telemetry instrumentation and validation | 0.85 |
| DevOps Agent | CI/CD pipeline automation | 0.85 |
Agent interaction pattern
Agents interact through the State Manifest, not directly with each other:
This indirect interaction pattern:
- Prevents agents from directly overriding each other's outputs
- Creates an immutable audit trail of every state transition
- Allows the Knowledge Agent to detect conflicts before they cause cascading failures
Context injection: what agents receive
The Knowledge Agent is responsible for Dynamic Context Injection — ensuring each agent receives only the context relevant to its current task.
At each phase gate, the Knowledge Agent constructs a minimal context payload containing:
- The active slice's
phase_data(links to intent, requirements, design artifacts) - The applicable Ubiquitous Language terms for the current domain area
- The Steering Rules relevant to the current phase
- Prior Reasoning Traces from upstream agents in the same slice
- Any open dissent notices or blockers for the slice
Everything else is withheld. A clean context window is as important as clean code.
Reasoning traces
Every agent emits a reasoning_trace.md alongside its artifact. The trace documents:
- Key assumptions made
- Requirements prioritized and why
- Alternatives considered and rejected
- Uncertainty factors (required when confidence < 0.95)
Phase gate sign-off requires the TL to acknowledge the Reasoning Trace — not just the artifact.
Navigate the catalog
Each agent has a dedicated page with full specification:
- Discovery Agent — behavioral slicing and assumptions-first discovery
- Spec Agent — EARS requirements generation
- Validation Agent — ambiguity detection and quality scoring
- Domain Agent — domain model building and maintenance
- Design Agent — architecture synthesis and ADRs
- Task Planning Agent — execution wave decomposition
- Implementation Agent — code generation orchestration
- QA Agent — spec coverage validation and test generation
- Refactor Agent — code quality maintenance
- Knowledge Agent — system memory, state, and learning