wb-flow

Appearance

/wbContext — Examples ​

Self-help. /wbContext is the command you run more than any other. This file is the cheat sheet for "what's actually happening when I type it."

Format: annotated transcripts, same as wbSetup_examples.

The core distinction: /wbSetup vs /wbContext ​

/wbSetup/wbContext
Creates context.md + dev.md for the first timeRefreshes context.md against current code
Run once per package, everRun every session, and after any non-trivial change
Can overwrite hand-authored rules (risky)Safe to re-run — diffs before writing
Focus: what is this package?Focus: does my stored understanding still match reality?

If you find yourself re-running /wbSetup, you probably want /wbContext --refresh instead.

Example 1 — /wbContext packages/wb-core (start of session, no changes since yesterday) ​

The cheapest, most common invocation. You just opened a fresh AI chat.

text
$ /wbContext packages/wb-core

[AI] Reading .agents/workflows/context.md (baseline)...
[AI] → 11 sections, dated 2026-04-18 (5 days old)
[AI] Reading .agents/workflows/dev.md (rules)...
[AI] → 14 rules, dated 2026-04-18
[AI] Scanning reports/20260418..20260423/ ...
[AI] → 2 audit reports, 1 debug report in range
[AI]
[AI] Checking drift...
[AI] package.json: unchanged
[AI] src/index.js exports: unchanged
[AI] vite.config.js outDir: unchanged
[AI]
[AI] ✅ No drift. context.md still accurate.
[AI] ✅ Ingested 3 recent reports; notable:
[AI] - 2026-04-22 audit flagged: extractSubObject tests incomplete
[AI] - 2026-04-21 debug report: __WBC_DEV__ leaking in dev build
[AI]
[AI] Ready. I know this package.

What's worth noticing:

  • When nothing changed, /wbContext is a 2-second operation. It's not regenerating context; it's verifying it and ingesting recent reports.
  • The 2 flagged items (incomplete tests, __WBC_DEV__ leak) are now in the AI's working memory for this session. Without /wbContext, the AI would have no idea.
  • If you skip this command and jump into "implement X," the AI will implement X but miss the leak regression risk.

Example 2 — /wbContext packages/wb-core (you've been editing all week) ​

Same command, different situation. You've made 30 commits since the last context refresh.

text
$ /wbContext packages/wb-core

[AI] Reading baseline context.md...
[AI] → dated 2026-04-15 (8 days old)
[AI]
[AI] Checking drift...
[AI] package.json: vue 3.4.0 → 3.5.2 (changed)
[AI] src/index.js: +2 new exports (changed)
[AI] WBCode, WBCodeSlot
[AI] src/utils/: +1 file (extractPath.js)
[AI] vite.config.js: outDir: dist → dist-dev (changed!)
[AI]
[AI] ⚠ Drift detected in 4 places.
[AI]
[AI] Questions:
[AI] 1. The vite outDir change from dist/ → dist-dev/:
[AI] is this intentional? package.json main: still says dist/.
[AI] (This is the footgun your dev.md warns about.)
[AI] 2. WBCodeSlot is a new export but not documented. Should I
[AI] add it to context.md's public API list?
[AI] 3. extractPath.js — new utility or should this be part of
[AI] extractSubObject?
[AI]
[AI] Waiting for your answers before rewriting context.md.

What's worth noticing:

  • The AI doesn't silently rewrite. Drift → questions. This is the right posture: your intent is privileged, not the AI's interpretation.
  • Question #1 is the whole point of dev.md — the rule "don't rename main without updating vite" fires here and catches a real regression in progress.
  • /wbContext without answering the questions is safe (it just reads); answering lets it update context.md with the new reality.

Example 3 — /wbContext packages/wb-core --focus="extractSubObject" (deep dive) ​

You're about to touch extractSubObject. You want the AI loaded up on just that function.

text
$ /wbContext packages/wb-core --focus="extractSubObject"

[AI] Running standard context check first... (see Example 1)
[AI] Baseline confirmed.
[AI]
[AI] Focus requested: extractSubObject
[AI] Checking for existing focused file...
[AI] .agents/workflows/context_extractSubObject.md (2026-04-10, 13 days)
[AI]
[AI] Diffing against current code:
[AI] Function signature: unchanged
[AI] Test coverage: unchanged
[AI] Call sites: 3 → 4 (new caller in @wbc/wb-press)
[AI]
[AI] ⚠ Open decision tracker: the array-handling question
[AI] (leaves vs. structural) is still UNRESOLVED.
[AI] No consumer has yet used the structural shape.
[AI] Signature-stability risk is low for now.
[AI]
[AI] Updating context_extractSubObject.md:
[AI] - Refreshed call site count
[AI] - Noted new wb-press consumer
[AI] - Open decision re-stated (unchanged)
[AI]
[AI] ✅ Done. Focused context ready. Non-focused context.md
[AI] untouched.

What's worth noticing:

  • Focused context is additive. --focus creates/updates a sidecar file, never rewrites the base context.md.
  • The "open decision tracker" is a pattern worth preserving: undecided architectural questions get parked explicitly, not silently resolved.
  • The new wb-press consumer is worth flagging in its own right — it means the next /wbRelease needs to verify wb-press still works.

Example 4 — /wbContext apps/wb-dataviewer/wbdataviewer2.wbc-ui.com (the demo app) ​

Context refresh on an app, not a package. Different output shape because apps have different concerns.

text
$ /wbContext apps/wb-dataviewer/wbdataviewer2.wbc-ui.com

[AI] Reading baseline context.md...
[AI] Reading dev.md...
[AI]
[AI] Drift check against:
[AI] package.json: unchanged
[AI] src/App.vue: +15 lines (new example)
[AI] src/examples/: +1 file (Example_wbPress.vue)
[AI] vite.config.js: unchanged
[AI] apiResponse_ usage: still caches fetched data ✅
[AI] :wbCode="false": still used on all demo instances ✅
[AI]
[AI] Reports scan (last 3 days):
[AI] - No test failures.
[AI] - No deploy failures.
[AI]
[AI] Question:
[AI] The new Example_wbPress.vue uses :wbCode="true" (not false).
[AI] dev.md rule says all examples default to false. Is this
[AI] intentional (demonstrating wbCode) or oversight?
[AI]
[AI] Waiting for answer before updating context.md.

What's worth noticing:

  • The AI catches a :wbCode="true" instance and asks. It doesn't auto-"fix" — that's a feature. You might be demonstrating the toggle on purpose.
  • The apiResponse_ caching pattern and :wbCode="false" default are both positive checks (✅). The AI verifies the convention is still followed, not just flags violations.
  • App context tracks different things than package context. Demo apps care about examples/, App.vue, rendering shape. Libraries care about src/index.js and exports.

Example 5 — /wbContext core2/ --scope=global (monorepo-level) ​

Rare. Expensive. Don't run this often.

text
$ /wbContext core2/ --scope=global

[AI] Reading core2/.agents/workflows/monorepo_rules.md...
[AI] → dated 2026-03-01 (53 days old!)
[AI]
[AI] Surveying 23 packages in packages/...
[AI] Surveying 8 apps in apps/...
[AI]
[AI] Systemic drift check:
[AI] Vue 3 across packages: 22/23 ✅ (wb-press on Vue 2: expected)
[AI] workspace:* for @wbc/* imports: 31/31 ✅
[AI] dist/ vs dist-dev/ alignment: 14/16 ⚠ (2 misaligned)
[AI] wb-core: dist-dev (should be dist)
[AI] wb-code: dist-dev (should be dist)
[AI] wbCode dev-only gating: 17/17 ✅
[AI] monorepo_rules.md freshness: STALE (>30 days)
[AI]
[AI] Recommendations:
[AI] 1. monorepo_rules.md needs a refresh pass.
[AI] 2. 2 packages have misaligned main vs. build output. Fix before
[AI] next /wbRelease.
[AI]
[AI] Monorepo-level context report written to:
[AI] reports/20260423/contexts/context_global_202604231700.md
[AI]
[AI] Per-package context.md files NOT modified (use /wbContext
[AI] <package> for that).

What's worth noticing:

  • Global context is a survey, not an update. It tells you which rules are being followed and which aren't. You decide what to fix.
  • 53-day-old monorepo_rules.md is a soft failure. No alarm, just a note. In 6 months, this file will be wrong in ways you don't notice until a /wbRelease breaks.
  • The 2 misaligned dist/dist-dev packages are pre-ship blockers. This report would catch them before they reach users.

The pattern across examples ​

/wbContext is always doing three things in order:

  1. Load baseline — read the stored context.md + dev.md.
  2. Check drift — diff the stored understanding against the current code.
  3. Ingest recent reports — pull in the last 1–3 days of reports/ into working memory.

The output shape depends on drift:

  • No drift → quick ✅, proceed.
  • Drift found → questions, wait for answers, then update.
  • Global scope → survey report, no per-package updates.

Pattern variation: --focus= adds a fourth step (update the focused sidecar file), --scope=global replaces step 2 with a systemic survey.

When /wbContext fails silently ​

Three failure modes worth knowing about:

  1. You skip it. The AI has no stored understanding. It'll make generic suggestions that ignore your conventions (propose Vue 3 syntax for wb-press, propose Chart.js alongside D3, etc.).
  2. The baseline is months old. /wbContext loads stale rules. The AI confidently follows them. You discover months later that dev.md forbids something you no longer do.
  3. You run it but don't answer the drift questions. Next session opens with the same drift because context.md was never updated. Fix: always answer, even if the answer is "don't update, keep old baseline — I'll fix the code instead."

Mitigation for all three: treat /wbContext as a required session-opener. Not optional.

Basic Usage ​

bash
# Standard command execution
/wbContext frontEnd/wbc-ui3/packages/

# Execution with explicit target ID filter
/wbContext deployement/apps/wb-jobs/ --id=1,2,3

Model Fallback Chain (.wb/bin/wbRun) ​

When executing CLI dispatches, /wbContext wraps binary calls in .wb/bin/wbRun to ensure output-error guarding:

bash
.wb/bin/wbRun claude -p --permission-mode auto "/wbContext target/" || .wb/bin/wbRun agy --model gemini-3.1-pro-high -p "/wbContext target/" || .wb/bin/wbRun opencode run -m opencode-go/deepseek-v4-pro "/wbContext target/"

Role Model Overrides (--planner, --worker, --validator, --mechanical) ​

You can override default models per role directly from the command line:

bash
# Override Worker and Validator models
/wbContext packages/ --wave=A --worker="DeepSeek V4 Pro,Kimi K3" --validator="Gemini 3.5 Pro"

# Override all 4 roles simultaneously
/wbContext apps/wb-jobs/ --wave=all --planner="Claude Opus 5" --worker="DeepSeek V4 Pro" --validator="Claude Sonnet 4.7" --mechanical="Gemini 3 Flash"

Autonomous Non-Interactive Execution (-y / --yes) ​

Run /wbContext in zero-touch print mode without stopping for manual decision prompts:

bash
.wb/bin/wbRun claude -p --permission-mode auto "/wbContext frontEnd/wbc-ui3/packages/ --wave --as='expert,steps' -y"