Command Flow Diagrams
Visual flows showing how Rebar commands chain together in practice.
Client Onboarding
flowchart LR A["/create my-client"] -->|generates config| B[client.yaml<br>expertise.yaml<br>notes.md] B --> C["/discover my-client"] C -->|pulls from all sources| D[phase-0-discovery.md<br>seeded expertise.yaml] D --> E["/check my-client"] E -->|validates against Phase 0| F[compliance report] F --> G["/brief my-client"] G -->|reads expertise.yaml| H[standup summary<br>suggested focus areas]
Start with /create to scaffold the client directory. /discover fills it with everything the framework can find. /check validates completeness. /brief gives you a starting point for the first session.
Development Cycle
flowchart TD A["/brief my-app"] -->|read context| B{What needs doing?} B -->|new feature| C["/plan feature description"] B -->|bug report| D["/bug my-app symptom"] B -->|investigation| E["/scout my-app question"] C -->|creates spec| F[specs/feature-name.md] F --> G["/build specs/feature-name.md"] G -->|implements plan| H[code changes] H --> I["/test my-app"] I -->|passes| J["/review my-app"] I -->|fails| K[fix failures] K --> I J --> CL["/close-loop feature-name"] CL --> EV[Evaluator validates] EV --> GATE{Release gate<br>scans for blocker language} GATE -->|blockers found| BLOCKED[Ticket BLOCKED<br>fix + retry] GATE -->|clean| IMP["/improve app --from eval-file"] IMP --> MI["/meta-improve<br>2+ occurrence patterns to queue"] MI --> APPLY["/meta-apply<br>human review"] APPLY --> WI["/wiki-ingest"] BLOCKED --> G D --> IMP E --> IMP WI -->|next session| A
Every feature ends with /close-loop — the harness that validates, gates on blocker language (“must generate migration,” “cannot ship”), promotes cycle-scoped observations, detects recurring failure patterns, surfaces template patches for human review, and ingests durable knowledge into the wiki. Bug fixes and investigations skip straight to /improve — they don’t need the full harness. See architecture.md → Close-Loop Harness for the full four-gate diagram.
Knowledge Capture
flowchart LR subgraph "Intake" R1[Meeting notes] R2[Web clips] R3[PDFs] R4[Transcripts] end subgraph "Processing" DROP[Drop in raw/] INGEST["/wiki-ingest"] FILE["/wiki-file topic"] end subgraph "Wiki" PAGES[wiki pages created] LINKS["[[cross-links]] added"] INDEX[index.md updated] LOG[log.md appended] end subgraph "Health" LINT["/wiki-lint"] FIX[auto-fix orphans<br>broken links] end R1 --> DROP R2 --> DROP R3 --> DROP R4 --> DROP DROP --> INGEST INGEST --> PAGES PAGES --> LINKS LINKS --> INDEX INDEX --> LOG FILE --> PAGES LINT --> FIX FIX --> LINKS
Two paths into the wiki: drop files in raw/ and run /wiki-ingest, or capture conversation insights with /wiki-file. Both produce linked, indexed wiki pages. /wiki-lint keeps everything connected.
Feature Workflow (Detailed)
flowchart TD START["/feature my-app add payment processing"] START --> LOAD[Load expertise.yaml<br>Read existing patterns] LOAD --> PLAN[Generate implementation plan<br>Save to specs/] PLAN --> REVIEW_PLAN{Review plan?} REVIEW_PLAN -->|approved| BUILD[Execute plan step-by-step] REVIEW_PLAN -->|edit| PLAN BUILD --> STEP1[Step 1: Create models] STEP1 --> STEP2[Step 2: Add routes] STEP2 --> STEP3[Step 3: Wire services] STEP3 --> STEP4[Step 4: Add tests] STEP4 --> TEST[Run test suite] TEST -->|pass| OBSERVE[Append observations] TEST -->|fail| FIX[Fix failures] FIX --> TEST OBSERVE --> UPDATE[Update expertise.yaml] UPDATE --> DONE[Feature complete]
The /feature command combines /plan and /build into a single workflow. It loads project context first, so the plan reflects existing patterns and known limitations.
Agent Coordination Flow
flowchart TD subgraph "Morning (8am)" GTM[GTM Agent] -->|pull metrics| METRICS[LinkedIn, Reddit, GitHub] GTM -->|identify| PROSPECTS[New prospects] GTM -->|decide| TOPIC[Today's topic + angle] end subgraph "Morning (9am)" SM[Social Media Agent] -->|reads| TOPIC SM -->|generates| DRAFT[Draft post] DRAFT --> HUMANIZE[Humanize via Claude] HUMANIZE --> POST[Post to LinkedIn + Facebook] POST --> SAVE[Save to system/drafts/] end subgraph "Every 30 min" OR[Outreach Agent] -->|checks| COMMENTS[New comments on posts] COMMENTS --> CLASSIFY{Classify} CLASSIFY -->|question| ANSWER[Generate answer] CLASSIFY -->|objection| EVIDENCE[Lead with evidence] CLASSIFY -->|compliment| THANKS[Thank + add value] ANSWER --> HUMANIZE2[Humanize reply] EVIDENCE --> HUMANIZE2 THANKS --> HUMANIZE2 HUMANIZE2 --> REPLY[Post reply] end subgraph "Every 5 min" TR[Triage Agent] -->|reads| ISSUES[Unassigned issues] ISSUES --> ROUTE{Route} ROUTE -->|wiki| WC[Wiki Curator] ROUTE -->|build| SB[Site Builder] ROUTE -->|expertise| CS[Rebar Steward] end
The GTM Agent sets the day’s strategy. The Social Media Agent executes it. The Outreach Agent handles engagement. The Triage Agent routes everything else.
Related
- Architecture — system-level diagrams
- Commands — full command reference
- Paperclip — agent setup