notque/vexjoy-agent / skills/meta/d
Router that runs the work in the main loop: picks skill, pipeline, and stack with /do's rules; Jev checks intent and attaches skills.
| File | Kind | Size |
|---|---|---|
| EVAL.md | markdown | 8.4 KB |
| SKILL.md | skill md | 15.1 KB |
| SPEC.md | markdown | 10.2 KB |
| references/jev-classifier-design.md | markdown | 11.6 KB |
| references/model-task-fit.md | markdown | 13.6 KB |
| references/router-metrics.md | markdown | 4.3 KB |
references/router-metrics.md
View on GitHub# Router metrics after agent retirement (plans/009 L6)
Load when comparing `/do` and `/d`, or before quoting a routing figure. Every
figure is a dated observation: rerun the command before quoting it.
## What each metric means now
The 44 specialist agents are retired. Every route runs on the main thread
(`agent=main`) or on a `general-purpose` worker, and the skill carries the
domain. Metrics that scored the agent pick lost their meaning.
| Metric | Source | Status | Why |
|---|---|---|---|
| `route-health` outcome coverage, basis split, silent-success share | `learnings`, `routing_outcome_basis` | Kept, fixed | Bookkeeping rows (`stack-usage:`, `fallback-reason:`) no longer count as routes; they never get an outcome and inflated the no-outcome share (43 of 133 rows on 2026-11-17). |
| Fallback rate | `fallback-reason:invalid-agent:*` rows | Kept (L5 redefined it) | Now counts coerced agent names only. Near 0 by design. |
| `general-purpose` share | `evidence_route_decisions.agent` | Retired as a quality signal | About 100% of dispatches by design. Read `agent` as main vs dispatch instead. |
| Skill concentration, skill entropy | `evidence_route_decisions.skill` | Kept | The skill is the routing choice that remains. |
| Route-fit verdicts | `routing_outcome_basis` `route_fit:*` | Kept for dispatches only | A main-loop route has no worker reply, so it has no verdict. `wrong-agent` is retired. Not comparable across routers with different dispatch shares. |
| `tool_errors` per route | recorder value token | Kept for dispatches only | Main-loop routes record no Agent result. Not comparable across routers. |
| `route-stats --by errors` | recorder value token | Fixed | It read ` | ` pairs the recorder stopped writing; it now reads `tool_errors=1`. |
| `route-stats --by force-route`, `--by override` | none | Retired (prints "not instrumented") | No writer has recorded `force_used` or `llm_override` in the current value shape: 0 rows. |
| `handoff-report` spec score | `evidence_route_decisions.spec_score` | Kept, fixed | Main-thread rows are excluded: their prompt is the bare marker, so they always scored 0 (6 of 6 rows). |
| Health / alts / action tokens | marker | Retired earlier (`docs/route-loop-validation.md`) | Unchanged. |
## Per-router metrics (new)
`python3 scripts/router-metrics.py [--since <date>] [--json]` reads
`route-events.jsonl` and `usage.db`, read-only, and splits by router. A `/d`
marker carries `router=d`; no token means `/do`. Before L6 `/d` emitted no
router token, so start comparisons at `--since` the install date.
| Metric | Definition |
|---|---|
| `main_share` / `dispatch_share` | decisions with `agent=main` / with a worker |
| `dispatch_reasons` | `/d` `reason=` token per dispatch (`fan-out`, `worktree`, `haiku-gather`, `verifier`, `overlay-agent`, `context-budget`); `/do` dispatches read `unstated` |
| `pipeline_rate` | decisions whose marker carries `pipeline=` |
| `verify` | `/d` main routes marked `verify=required`, and how many saw a `reason=verifier` dispatch later in the same session |
| `skill_load.main` | main-loop decisions where the session called the Skill tool with the routed skill within -2 to +30 minutes |
| `skill_load.dispatch` | same for dispatches; a lower bound, because a worker's Skill calls mostly do not land under the parent session (29 of 329 observed on 2026-11-17) |
Compare `skill_load.main` across routers, not the pooled rate: a main-heavy
router would win the pooled number on observability alone.
## Golden eval
`scripts/golden_eval/` holds 14 tasks shaped on the owner's routed work, each
with acceptable skills and pipelines, required skill groups, forbidden skills,
allowed `/d` dispatch reasons, a rubric, and (for 6 tasks) acceptance checks on
a fixture repo.
| Level | Command | Cost |
|---|---|---|
| Routing (each task twice per router, same model) | `nice -n 10 python3 scripts/golden_eval/run_routing.py --out-dir scripts/routing-ab-results/<new> --runs 2` | about $0.2 per routing call on opus, about 1k Jev tokens per `/d` call |
| End to end (real sessions, acceptance checks, rubric grade, skill-load from the stream) | `nice -n 10 python3 scripts/golden_eval/run_e2e.py --out-dir <new> --runs 2 --launch` | `run_e2e.py --estimate`; needs this branch installed |
Count a task as changed between arms only when both runs of each arm agree.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |