Command Composition: Self-Application & Chaining
Audience: Developers who want to chain
/wb*commands or feed a command's output back into itself. Related: Command Classification · wbPlan Flags · Overview
Why this matters
Most /wb* commands produce a markdown file; many of those files can be fed back into the same or a different /wb* command. This page tells you which chains make sense and which don't.
Prerequisite: Command Classification — you need to know what each command produces before you can chain them.
Notation
Two equivalent forms:
| Form | Example | How to read it |
|---|---|---|
| Pipe form (recommended) | /wbNext folder/ | /wbPlan | "Run /wbNext on folder/, then feed the output file to /wbPlan" |
| Nested form | /wbPlan /wbNext folder/ | Same meaning — inner command's output feeds the outer command |
Use parentheses when flags make things ambiguous: /wbPlan(/wbNext folder/ --recent). Otherwise the bare nested form is fine.
Three ways a command "feeds itself"
| Pattern | What it means | Example |
|---|---|---|
| Temporal memory (implicit) | A single invocation reads its own past outputs from reports/ before writing. Automatic. | /wbAudit Smart-Merge appends Entry #N |
| Self-application (explicit) | Two invocations of the SAME command — second one takes the first's output file as input. | /wbAudit audit_*.md — audit the audit |
| Chaining (explicit) | Two invocations of DIFFERENT commands — second consumes first's output file. | /wbNext folder/ | /wbPlan |
- Temporal memory happens automatically. You don't invoke anything twice.
- Self-application is deliberate: you point a command at its own report.
- Chaining moves a work artifact from one lifecycle stage to the next.
Self-application table — all 30 commands
Can each command run on its own output file? What happens if it does?
Outcome classes:
- 🟢 Meaningful — second run produces real new value
- 🟡 No-op — type-valid but produces nothing useful
- 🔴 Cannot — command's invariants forbid it
| # | Command | Can self-apply? | What the second run does |
|---|---|---|---|
| 01 | /wbStandup | 🟡 No-op | Summarizing a summary yields nothing new |
| 02 | /wbPlan | 🟢 Meaningful | Re-plans an existing plan (--resume mode appends tasks) |
| 03 | /wbNext | 🟡 No-op | Ranking a ranked recommendation adds nothing |
| 04 | /wbVision | 🟡 No-op | Proposing features from a proposal = meta-features |
| 05 | /wbExplain | 🟢 Meaningful | Explains an existing explanation at a different layer (ELI5 vs Expert) |
| 06 | /wbAudit | 🟢 Meaningful | Cross-model audit-of-the-audit (verifies auditor's work) |
| 07 | /wbSecure | 🟡 No-op | Scanning a markdown report for code vulns finds nothing |
| 08 | /wbDebug | 🟡 No-op | Debugging a debug report won't fix the code |
| 09 | /wbReview | 🟢 Meaningful | Reviews a prior review — sanity-checks the reviewer |
| 10 | /wbValid | 🟢 Meaningful | Cumulative multi-model QA (appends validation signature) |
| 11 | /wbTest | 🟡 No-op | Test reports aren't executable test suites |
| 12 | /wbCheck | 🔴 Cannot | Console output only; no persistent file |
| 13 | /wbWork | 🔴 Cannot | Once done, re-running on the task report doesn't re-execute |
| 14 | /wbRefactor | 🔴 Cannot | Requires source code, not markdown reports |
| 15 | /wbClean | 🟡 No-op | "Cleaning the cleanup report" — type-valid, useless |
| 16 | /wbTranslate | 🔴 Cannot | Requires UI components, not reports |
| 17 | /wbToWBC | 🔴 Cannot | Requires legacy HTML/Vuetify, not migration reports |
| 18 | /wbSetup | 🔴 Cannot | Once-only bootstrap; re-running on output is a no-op |
| 19 | /wbContext | 🟢 Meaningful | Refreshes an existing context file with current state |
| 20 | /wbDoc | 🟡 No-op | Generates docs about docs — semantically thin |
| 21 | /wbHelp | 🔴 Cannot | Console output only; no file produced |
| 22 | /wbRelease | 🔴 Cannot | Requires codebase state, not release logs |
| 23 | /wbPublish | 🔴 Cannot | Publish log records a past NPM push; can't re-publish |
| 24 | /wbDeploy | 🔴 Cannot | Deploy log describes a past event, not a deployable artifact |
| 25 | /wbBroadcast | 🟡 No-op | Announcing an announcement kit = no new reach |
| 26 | /wbGit | 🔴 Cannot | Console only; commit string has no persistent file |
| 27 | /wbMonetize | 🔴 Cannot | Scaffolds in package.json; can't run on its own report |
| 28 | /wbLicense | 🔴 Cannot | Injects code into components; needs source files |
| 29 | /wbActOn | 🟢 Meaningful | Re-ranks with different priority criteria |
| 30 | /wbTrack | 🟢 Meaningful | Appends to itself natively (temporal memory per session) |
The core rule: stateless vs stateful
| Category | Commands | Rule |
|---|---|---|
| Stateless analyses | /wbAudit, /wbReview, /wbExplain, /wbContext, /wbDoc, /wbVision, /wbDebug, /wbSecure, /wbTest, /wbStandup, /wbNext | They observe; they don't advance any lifecycle. Almost always self-applicable (🟢 or 🟡). |
| Stateful transitions | /wbDeploy, /wbPublish, /wbWork, /wbSetup, /wbMonetize, /wbLicense, /wbRelease, /wbRefactor, /wbClean, /wbToWBC | They advance an artifact along a lifecycle (built→deployed). The output records the post-state. Re-running on the post-state is a category error. |
You cannot "deploy what is already deployed." The deploy log is a receipt of a past event, not a deployable artifact.
Chaining — useful command pairs
~10 chains that provide real workflow value. Not all 900 possible pairs.
| Chain (pipe form) | Nested form | What it does | Already wired as a flag? |
|---|---|---|---|
/wbNext folder/ | /wbPlan | /wbPlan /wbNext folder/ | Turn next-recommendation into a plan | No — manual chain |
/wbAudit folder/ | /wbActOn | /wbActOn /wbAudit folder/ | Rank audit findings by severity | Yes — /wbAudit --act |
/wbReview --plan=plan.md | /wbActOn | /wbActOn /wbReview --plan=plan.md | Rank review findings | Yes — /wbReview --act |
/wbAudit folder/ | /wbPlan | /wbPlan /wbAudit folder/ | Build plan from audit findings | Partial — /wbAudit --wbPlan |
/wbStandup monorepo/ | /wbActOn | /wbActOn /wbStandup monorepo/ | Rank standup items by priority | Yes — /wbStandup --act |
/wbVision folder/ | /wbPlan | /wbPlan /wbVision folder/ | Plan one of the vision proposals | No — manual chain |
/wbExplain plan.md --id=N | /wbCheck | /wbCheck /wbExplain plan.md --id=N | Pre-flight quiz the worker before execution | No — manual chain |
/wbAudit folder/ | /wbAudit | /wbAudit /wbAudit folder/ | Cross-model audit-of-the-audit | No — manual chain |
/wbRelease monorepo/ | /wbBroadcast | /wbBroadcast /wbRelease monorepo/ | Generate announcement from release | Yes — /wbBroadcast reads release logs |
/wbAudit folder/ | /wbActOn | /wbPlan | /wbPlan /wbActOn /wbAudit folder/ | Three-step: audit → rank → plan | Yes — /wbAudit --act --wbPlan |
/wbAudit packages/wb-core/ --act --wbPlanWorked examples
Example 1: The 3-Step Chain
/wbAudit folder/ | /wbActOn | /wbPlan
/wbAuditproducesaudit_*.md./wbActOnconsumes the audit →action_audit_*.md(ranked list)./wbPlanconverts the ranked list →plan_action_*.md(task table).
Shortcut: /wbAudit folder/ --act --wbPlan does the same thing in one invocation.
Example 2: Meaningful Self-Application
/wbAudit /wbAudit folder/
You ran an audit on Tuesday using the agent the agent. On Wednesday, you run /wbAudit again on the audit_*.md file using the agent Pro. Cross-model "audit of the audit" catches blind spots. New findings append as Entry #2.
Example 3: Chaining with an Explicit Plan
/wbPlan /wbNext folder/
/wbNext reads codebase + reports → next_*.md recommending a new API endpoint. /wbPlan reads that recommendation → structured worker plan with task IDs.
Counter-example: Stateful Violation
/wbDeploy /wbDeploy app/
Why it fails: app/ is "built but not deployed." First /wbDeploy → "deployed" + deploy_<date>.md (URL/logs). Feeding deploy_<date>.md back to /wbDeploy fails: the file describes a past event, not a deployable artifact. The command needs an undeployed input. This is a state-machine violation.
Quick reference — common chains
| I want to... | Chain | Notes |
|---|---|---|
| turn audit findings into a TODO list | /wbAudit folder/ | /wbActOn | or use /wbAudit --act |
| turn audit findings into a full plan | /wbAudit folder/ | /wbActOn | /wbPlan | or use /wbAudit --act --wbPlan |
| second-pass an audit by a different model | /wbAudit audit_*.md | smart-merge appends Entry #N |
| extend an existing plan | /wbPlan plan_*.md | triggers --resume mode |
| turn a next-rec into a plan | /wbPlan /wbNext folder/ | manual chain |
| verify a completed plan | /wbReview --plan=plan_*.md folder/ | adversarial check |
| rank standup agenda | /wbStandup monorepo/ | /wbActOn | or use /wbStandup --act |
- — companion to Command Classification.*