Work scope inventory
Skill lubochka/xiigen-mvp-engine/.claude/skills/universal/work-scope-inventory
Self-building AI code generation engine that generates application flows instead of implementing them. AGPL-3.0.
npx -y skills add lubochka/xiigen-mvp-engine --skill work-scope-inventoryAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 14 days oldThe repository was created 14 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.0 KB, 831 tokens by cl100k_base, as published. Nobody here has run it
Work Scope Inventory
Purpose
Use this to coordinate constrained workstreams, parallel agents, and tasks with strict file boundaries.
When to Use
Invoke when the user names allowed paths, excluded areas, workstream ownership, no-git rules, or concurrent writers.
Actions
- List allowed paths exactly and normalize them before use.
- List prohibited paths and operations.
- Identify expected artifacts and any index or registry updates required inside scope.
- Check existing files in scope so generated work does not collide accidentally.
- Before final response, compare changed files against the inventory.
XIIGen Adaptation
- For XIIGen workstreams, keep runtime code, root docs, docs/ai-skills, or other agent-owned areas out of scope unless explicitly allowed.
- If the scope includes only skills docs, verify only skill paths and skill indexes.
Avoid
- Do not rely on memory for path boundaries.
- Do not create helper files outside the allowed tree.
- Do not import source architecture, algorithms, design patterns, training, DPO, source domains, source classes, or source-specific paths.
Completion Signal
- The active work is limited to the authorized paths and every required in-scope artifact is accounted for.
G08 universal addition from llm_mvp_core — the DENOMINATOR, the 6-count protocol, Claim Authority
The path-boundary guidance above is one half of work-scope-inventory. The other, stronger
half from the core standard is the denominator discipline: establish how many total units
exist BEFORE the first edit, so a "pipeline improved" claim cannot be made from a handful of
touched files. This is the universal home (it pairs with the mvp coverage-completeness gate).
1. Establish the denominator before the first line
scope tracking record (write before any edit):
totalSubprojects : count of ALL units in the relevant tree (the denominator)
complete : units green by their gate + evidence
partial : started, not green
blocked : cannot start (missing prerequisite)
notStarted : remainder
thisSessionCovers : the units THIS session touches (the numerator)
For mvp a "subproject/unit" is a NestJS module, an engine node-handler, a React feature, or a
RAG sidecar route. totalSubprojects comes from the monorepo root index / actual tree
enumeration (NOT a docs/SUBPROJECT-MAP.md — mvp does not use that file for the count).
2. The 6-count protocol
Count, this session, with real commands:
1 total units (e.g. enumerate server/src modules + node-handlers + client features + rag routes)
2 complete (npx jest per unit → 0 failed, + DNA/contract evidence)
3 partial
4 blocked (prerequisite-chain BLOCKING)
5 notStarted
6 thisSessionCovers (the numerator)
3. Claim Authority table
WITHOUT a scope record → only UNIT-LEVEL claims allowed ("module X now passes").
No pipeline/overall claim permitted.
WITH a record, numerator <20% → still unit-level; a "pipeline improved" claim is NOT authorized.
WITH a record, numerator ≥20% → pipeline-level claim allowed ONLY after the
coverage-completeness gate confirms the denominator and the
remaining counts. A bare "X% of files" is not authority.
A "the pipeline is better" claim with no denominator is the exact failure this gate stops.
Integrates with coverage-completeness-gate and goal-delivery-completeness.
Note-only (NOT ported — stays in G12, R5)
Scope accounting for shared-model training sets (DPO row coverage) belongs to llm_mvp_core;
here the denominator counts product units only.