Skip to content

notque/vexjoy-agent / skills/meta/do

Classify user requests and route to the correct skill, pipeline, and stack. Primary entry point for all delegated work.

references/lazy-completion-detector.md

View on GitHub
<!-- pairs_with: do, quality-loop -->

# Lazy-Completion Detector

A check the orchestrator runs against an agent's "done" claim before accepting it. Scope: the **agentic-laziness** failure mode ONLY.

**Terminology:** "workflow" is canonical; "pipeline" is the legacy alias for the same concept.

## Scope (read first)

This file fills ONE gap: laziness — an agent stops at partial progress and declares done. The other two agentic failure modes are already covered elsewhere; this file does NOT re-cover them:

- **Goal drift** → `skills/meta/do/references/quality-loop.md` PHASE 5 (INTENT VERIFY): adversarial check that the diff matches the original request.
- **Self-preferential bias** → quality-loop PHASE 7 (FIX): use independent context when a specific blind spot warrants it; fresh-agent fixing is not mandatory.

Do not duplicate those controls here.

## The Laziness Failure

An agent reports success after partial work: "fixed 20 of 50 items," then declares done. Tests on the 20 pass, so PHASE 3 stays green and the gap survives.

## The Check

After an agent returns and before declaring the task done, compare **claimed scope** against **objective scope**:

| Step | Action |
|---|---|
| 1. Extract objective count | From the request: how many items/files/cases were in scope? ("all 50 lint errors", "every endpoint") |
| 2. Extract claimed count | From the agent's report and diff: how many did it actually touch? |
| 3. Compare | claimed < objective → INCOMPLETE, not done |
| 4. Re-dispatch | Fix: send a fresh agent the REMAINDER (objective minus claimed) with the same skill stack; re-run the check until claimed == objective |

Reject the early false-positive "done" at evaluation time. A passing test on a subset is not completion of the whole.

## When to Escalate

- Objective scope is enumerable (a count, a file list, "all X") and the claim covers less → re-dispatch the remainder.
- Objective scope is open-ended (no countable target) → fall back to quality-loop PHASE 5 intent-verify; this detector does not apply.

## Wiring

Run at `/do` Phase 4 EXECUTE evaluation, after an agent returns, before "done." Inside the quality-loop, run alongside PHASE 8 RETEST.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 119kUpdated Claude CodeCodexVerifiedInstall
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: 118kUpdated Claude CodeCodexVerifiedInstall