/wbActOn — Exhaustive Simulation ()
/wbActOn is the triage officer. It takes a diagnostic document (an audit, plan, review, standup, or any structured markdown) and converts it into a ranked execution order. The central design principle: annotate, never author. /wbActOn doesn't invent findings — it reads the source, applies a 5-color decision tree to every section, and produces an ordered thread of what to do first.
Read this if you want to know how the 5-color system works, why the --wbPlan flag exists separately from /wbPlan, and what "Smart Merge" means when multiple models triage the same source.
1. Role & target
| Aspect | Behavior |
|---|---|
| Role | The Triage Officer. Ranks and annotates; never invents. |
| Target | Any markdown file — typically a /wb* report output, but works on third-party docs, design memos, competitor analyses. |
| Cell scope | None. /wbActOn never touches plan cells or source code. |
| Side effects allowed | Writing the action file. With --wbPlan: also writing a companion plan file. |
| Side effects forbidden | Modifying the source document. Editing code. Mutating plan state. |
The "annotate, never author" rule is the hardest constraint. When the source says "there's a caching bug," /wbActOn annotates it with 🔴 /wbDebug "apiResponse_ cache invalidation" — it doesn't describe the bug itself, and it doesn't propose a fix. It points to the command that would fix it. The user decides whether to follow.
2. Argument resolution
| Form | Example | What /wbActOn does |
|---|---|---|
| File path (under reports/) | Command: /wbActOn audit_wb-core_20260504.md | Mirror mode. Reads the audit, writes action file to the same target's reports/<date>/actions/. |
| File path (elsewhere) | Command: /wbActOn competitor_analysis.md | Side-car mode. Anchors to monorepo root: core2/.agents/workflows/reports/<date>/actions/. |
| Folder path | Command: /wbActOn packages/wb-core | Pick mode. Lists recent *.md under the folder's reports/, asks which to process. If only one exists, proceeds without asking. |
The three modes are detected from input type, not from a flag. File → process directly. Folder → pick from candidates. This keeps the invocation simple while handling the common case (act on the latest audit) and the edge case (act on an arbitrary document).
3. Flag matrix
| Flag | Shortcut | Purpose |
|---|---|---|
--act | -a | On upstream commands (/wbAudit, /wbReview, /wbStandup): chain the command's output through /wbActOn automatically. |
--wbPlan | -P | Also produce a companion plan file with one task table per 🔵 finding. |
The flags are independent and composable:
| Combination | Result |
|---|---|
/wbAudit packages/wb-core | Audit only |
/wbAudit packages/wb-core --act | Audit + action file |
/wbAudit packages/wb-core --wbPlan | Audit + plan file |
/wbAudit packages/wb-core --act --wbPlan | Audit + action + plan (chain of three) |
Why both flags instead of one? --act alone is the cheap path — triage without committing to a multi-section plan. --wbPlan alone skips triage when you already know what needs planning. Together is the full chain. The separation matches how work actually flows: sometimes you just want to prioritize, sometimes you want a plan, sometimes both.
4. The 5-color decision tree
This is the core engine. For every finding in the source, /wbActOn walks this tree:
| Decision | Color | Output |
|---|---|---|
| Console is red? (error, failing build) | 🔴 | /wbDebug "<exact error message>" |
| Work fits one sentence? | 🟢 | Exact one-shot prompt to paste |
| File-level mess? (>1000 LOC, dead code) | 🟡 | /wbClean <pkg> then /wbRefactor <file> |
| Strategic/business decision? | 🟣 | No command. Human decides; agent advises. |
| Multi-step coordinated work? | 🔵 | /wbPlan <target> with goal/constraints/scope |
Each color also carries a recommended model:
| Color | Default model | Why |
|---|---|---|
| 🟢 (simple) | the agent 4 / Qwen3 Coder | Fast, cheap, sufficient |
| 🟢 (subtle) | the agent 4 | Correctness on tricky one-liners |
| 🔴 | the agent 4 | Root-cause analysis needs reasoning |
| 🟡 (sweep) | Qwen3 Coder / the agent Flash | Repetitive, high-throughput |
| 🟡 (refactor) | the agent 4 | Cross-file judgment |
| 🟣 | the agent 4 / the AI agent | Long-form synthesis |
| 🔵 | the agent 4 | Coordination + dependency reasoning |
The recommendation is advisory — the user can override. But it's always stated explicitly. "You decide" is not acceptable.
5. Pipelines (the agent-native scenarios)
/wbActOn audit_wb-core_20260504.md> /wbActOn audit_wb-core_20260504.md💠 Pipeline Post-audit triage on wb-core
The canonical use case. /wbAudit produced 23 findings. Now rank them:
💠 Pipeline Chained audit with plan generation
The full chain: audit → triage → plan, in one invocation:
💠 Pipeline Multi-model Smart Merge
Two models triage the same source. The second model's run triggers Smart Merge:
6. Edge cases & refusals
| Trigger | What /wbActOn does |
|---|---|
| Source has no structure (free-form notes) | Halt. ❌ Cannot triage unstructured content. Fix the source — /wbActOn needs sections, tables, or numbered items. |
| Source has no actionable findings (pure state report) | §0 says: ℹ️ No actions — this is a state snapshot. Use it as input to a future command. |
--wbPlan but zero 🔵 findings | ℹ️ No multi-step findings to plan. All actions are inline (🟢) or immediate (🔴/🟡). |
| Callout says "you should refactor this" without the command | Self-correction: always include the exact prompt or /wb* command. A callout without a pasteable command violates the hard rules. |
| Source explicitly says something is working and shouldn't be touched | "What I would NOT touch" section lists it. Prevents reactive over-refactoring. |
/wbActOn plan_wb-core_20260504.md --wbPlan | ❌ Error: source is already a plan. Use --act alone to re-rank it. |
The unifying principle: /wbActOn is the theory-to-practice bridge. It takes a diagnostic document and produces an execution order. It never invents findings, never modifies the source, always justifies the rank, and always states which model should do the work. The user acts on the output — the triage officer just sets the priorities.