Orchestrata Ultimate › Plans › agent-implementation-plan
Agent Implementation — Pipeline Automation
Source: plans/agent-implementation-plan.md at 8a2cb42ad · last changed 2026-07-27 · first committed 2026-06-26 · 106 lines.
Prose pending.
Plan text
Agent Implementation — Pipeline Automation
Branch: deepseek-edition Specs: SPEC-103 (Tool-Calling), SPEC-019 (Agents) Status: Phase 4 — Observer Logging
Agent Roster
| # | Agent | Alias | LOC | LLM Access | Tool Access | Pipeline Role |
|---|---|---|---|---|---|---|
| 1 | ConciergeVolk | U:Cx | 444 | Yes (call_llm) | search, ast_query | Intake — elicits requirements, detects work type |
| 2 | AnalystAgent | U:BA | 336 | Yes (call_llm) | search, ast_query, read | Analysis — requirements decomposition, story generation |
| 3 | PMVolk | U:PM | 338 | Yes (call_llm_with_tools) | ast_query, read_file | Scoping — estimates, warnings |
| 4 | ArchitectVolk | U:Ar | 342 | Yes (constitution-aware) | read, search, ast_query, list_dir | Design — architecture, file tree, API spec |
| 5 | CodeDirectorVolk | U:Cd | 391 | Yes (call_llm_with_tools) | read, write, list_dir, search, ast_query, run_shell | Implementation — code generation, swarm delegation |
| 6 | QaVolk | U:Qa | 3,133 | Yes (constitution-aware) | read, write, run_shell, search, check_coverage | QA — test generation, code review, auto-fix |
| 7 | ComplianceVolk | U:Cp | 443 | No (deterministic) | — | Traceability — RTM generation, design-vs-code |
| 8 | DocumentorVolk | U:Do | 163 | Yes (call_llm_with_tools) | read, write, list_dir, search, ast_query | Documentation — readme, specs, architecture, flow, API docs |
| 9 | JuniorVolk | J:Junior | 3,228 | Yes (call_llm) | read, write, run_shell, list_dir, search | Parallel — design review, code gen, dep validation |
| 10 | UXVolk | U:Ux | 241 | Yes (call_llm_with_tools) | read_file, search_code | UI Design — wireframes, design tokens, accessibility |
| 11 | SwarmCoordinator | — | 968 | Yes (dispatches juniors) | — | Parallel dispatch — isolated workspaces, atomic merge |
Tool Access by Agent
| Agent | read_file | write_file | list_dir | search_code | run_shell | ast_query | check_cov |
|---|---|---|---|---|---|---|---|
| U:Cx | — | — | — | ✓ | — | ✓ | — |
| U:BA | ✓ | — | — | ✓ | — | ✓ | — |
| U:PM | ✓ | — | — | — | — | ✓ | — |
| U:Ar | ✓ | — | ✓ | ✓ | — | ✓ | — |
| U:Cd | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — |
| U:Qa | ✓ | ✓ | — | ✓ | ✓ | — | ✓ |
| U:Cp | — | — | — | — | — | — | — |
| U:Do | ✓ | ✓ | ✓ | ✓ | — | ✓ | — |
| J:Junior | ✓ | ✓ | ✓ | ✓ | ✓ | — | — |
| U:Ux | ✓ | — | — | ✓ | — | — | — |
Work Type → Agent Pipeline
| Work Type | Stages | Description |
|---|---|---|
| feature | U:Cx → U:BA → U:PM → U:Ar → U:Cd → U:Qa → U:Cp | Full pipeline |
| refactor | U:Ar → U:Cd → U:Qa → U:Cp | Skip intake, redesign + fix |
| test | U:Cd → U:Qa → U:Cp | Generate tests, review |
| documentation | U:Cx → U:Do → U:Cp | Elicit context → doc agent |
| analysis | U:Cx → U:BA → U:Cp | Requirements analysis only |
| fix | U:Qa → U:Cd → U:Qa → U:Cp | Fix issues, re-review |
| review | U:Qa → U:Cp | Code review only |
| spec | U:Cx → U:BA → U:Do → U:Cp | New specification |
| dependency | U:Ar → U:Cd → U:Qa → U:Cp | Dependency updates |
| performance | U:Qa → U:Ar → U:Cd → U:Qa → U:Cp | Profile, optimize, verify |
Communication Architecture
PipelineRunner
│
│ _register_handlers(stages)
│ postman._handlers["U:Cx"] = handler_fn
│ postman._handlers["U:BA"] = handler_fn
│ ...
│
│ _run_stage(stage_spec, packet)
│ postman.route_sync(packet)
│ │
│ ├── Deterministic message ID (based on thread_id + content hash)
│ ├── Redis persistence (packet audit trail)
│ ├── Deduplication check
│ ├── CXPAuditor logging
│ ├── _dispatch_sync() → handler(packet)
│ │ agent = cls()
│ │ return await agent.process(packet)
│ └── Message log entry (CXPFeed visibility)
│
│ result = StageResult(status, output_packet, artifacts, duration_ms)
│
▼
PipelineResult (thread_id, stages, artifacts, metrics, overall_status)
Files Created/Modified
| File | Status | LOC | Purpose |
|---|---|---|---|
agents/tool_registry.py |
NEW | 390 | 7 tool definitions, ToolRegistry, Agent tool sets |
agents/volks/base.py |
MODIFIED | +65 | _call_llm_with_tools() — multi-turn tool-calling loop |
agents/volks/codedirector/prompts.py |
NEW | 57 | CODEDIRECTOR_SYSTEM_PROMPT with tools |
agents/volks/concierge/volk.py |
MODIFIED | +81 | WORK_TYPE detection + routing, observer logging |
agents/volks/documentor/volk.py |
NEW | 163 | Documentor agent — 6 doc types, tool access |
agents/volks/documentor/prompts.py |
NEW | 56 | DOCUMENTOR_SYSTEM_PROMPT with tools |
orchestrator/pipeline_router.py |
NEW | 274 | 10 work types → 9 capability → agent mapping |
orchestrator/pipeline_runner.py |
NEW | 309 | Postman-routed pipeline executor |
orchestrator/artifact_store.py |
NEW | 91 | Thread-safe artifact collection |
Verification
# All imports resolve
python -c "from orchestrata_ultimate.agents.volks.documentor import DocumentorVolk; from orchestrata_ultimate.orchestrator.pipeline_runner import PipelineRunner; print('OK')"
# Agent class resolution
python -c "from orchestrata_ultimate.orchestrator.pipeline_router import PipelineRouter; r=PipelineRouter(); [print(f'{wt.value}: {[s.alias for s in r.route(wt)]}') for wt in r.route.__func__.__defaults__[0].__class__]"
# Smoke test (review work type = QA → Cp only, fastest path)
python scripts/pipeline_quick_test.py