Skip to content

Latest commit

 

History

History
131 lines (90 loc) · 6.52 KB

File metadata and controls

131 lines (90 loc) · 6.52 KB

Evaluation: liza-mas Token-Saving CLI Tools

Source: GitHub org liza-mas (Tangi Vass) Context: Mentioned in a LinkedIn comment on Soufiane Keli's series on LLM token-cost optimization for technical teams Evaluated: 2026-06-10 Evaluator: Claude Opus 4.8


Executive Summary

Tool Stars Language License Score Decision
scip-search 0 Go Apache-2.0 1/5 Reject
mdtoc 1 Go Apache-2.0 1/5 Reject
functional-clusters 0 Go Apache-2.0 1/5 Reject
stacklit-cli 2 Go MIT 2/5 Watch (one-line note only)

Roster score: 1/5: No tool clears the integration gate (score >= 3 required). stacklit-cli earns a single-line watch note as a Go-native variant of the already-documented stacklit.


Tool-by-Tool Analysis

scip-search

Repo: github.com/liza-mas/scip-search Version: v0.1.0 (2026-06-10, released same day as this evaluation) Description: SCIP-based code navigation tool for AI agents. Works with Stacklit and Functional Clusters.

Criterion Assessment
Traction 0 stars, v0.1.0, single author, day-one release
Novelty vs. guide Overlaps with Serena (LSP-based symbol navigation), grepai (semantic search), and any LSP-aware editor tooling
CC integration No documented Claude Code integration beyond the Liza framework ecosystem
Claim verifiability No benchmarks, no independent usage reports
Dependency Requires SCIP index generated by Stacklit, not standalone

Score: 1/5: Too early to evaluate on merit. Zero external traction, tightly coupled to the Liza framework, and the core use case (navigating code symbols for AI agents) is already addressed by Serena and grepai in the guide.


mdtoc

Repo: github.com/liza-mas/mdtoc Updated: 2026-06-07 Description: Markdown parser generating a table of content with line ranges.

Criterion Assessment
Traction 1 star, no forks, single author
Novelty vs. guide Generic utility. Dozens of equivalent tools exist (doctoc, markdown-toc, github-markdown-toc). Not CC-specific.
CC integration None: no CLAUDE.md template, no hook, no slash command
Claimed reduction Not stated
Dependency Standalone, but solves a problem Claude Code handles natively (agent reads file structure)

Score: 1/5: A useful personal utility, but out of scope for the guide. Claude Code users do not need a separate TOC generator; agents read structure via Read or ctx_read. No CC-specific value.


functional-clusters

Repo: github.com/liza-mas/functional-clusters Updated: 2026-06-08 Description: Functional Clustering from SCIP and Stacklit indexes.

Criterion Assessment
Traction 0 stars, single author
Novelty vs. guide Clustering functions by call graph. Overlaps with grepai trace_graph and Serena symbol navigation
CC integration None documented
Dependency Requires both SCIP and Stacklit indexes to run
Standalone value Cannot be evaluated in isolation; depends on two other liza-mas tools

Score: 1/5: Framework-internal component rather than a standalone tool. Requires two upstream dependencies (scip-search, stacklit-cli) that themselves have zero traction. No evidence of usage outside the Liza framework.


stacklit-cli

Repo: github.com/liza-mas/stacklit-cli Updated: 2026-06-09 Description: One command gives AI agents instant codebase context. ~250 tokens replaces 50,000+ tokens of exploration. Auto-configures Claude Code, Cursor, Aider.

Criterion Assessment
Traction 2 stars, very recent
Novelty vs. guide The documented stacklit (glincker/stacklit) solves the identical problem ("~250 tokens replaces 50,000+ tokens") via npm. stacklit-cli is a Go reimplementation.
Differentiator Go binary with no npm/Node runtime dependency. Different author, different org.
CC integration Claims auto-configure Claude Code via stacklit setup equivalent
License MIT (documented stacklit: not clearly listed in repo)

Score: 2/5: Genuine concept (already documented as stacklit), different implementation runtime (Go vs npm). For teams on Go-native stacks that want to avoid Node, this is worth knowing. But it is not additive to the guide as a new entry. The concept and pattern are covered. Two stars is not enough traction to justify a separate tool section.

Watch-list note: A one-line variant mention under the existing stacklit entry is appropriate if traction grows. Do not create a new ### stacklit-cli section.


Challenge Assessment

Challenge agent position: Are the scores too harsh on stacklit-cli?

The documented stacklit (glincker) has its own low-traction concerns (the guide does not show its star count). stacklit-cli from liza-mas claims auto-configuration for Claude Code and targets a real pain point (codebase onboarding tokens). With 2 stars on a tool released in the same week, the traction argument against it is the same argument that would disqualify stacklit itself.

Response: The difference is that stacklit (glincker) was evaluated and documented before this policy was applied strictly; the guide has already committed to that entry. For any new entry, the bar is score >= 3. stacklit-cli reimplements an already-documented concept without demonstrating adoption. If stacklit-cli grows to 50+ stars and shows distinct production usage, re-evaluate. Until then, a one-line variant note is the proportionate response, not a full section.

Score maintained: stacklit-cli 2/5.


Integration Decisions

Tool Action Rationale
scip-search Reject 0 stars, day-one release, Liza-framework specific
mdtoc Reject Out of scope, generic utility, no CC value
functional-clusters Reject 0 stars, multi-dependency, framework-internal
stacklit-cli Watch only (no guide entry) Go variant of documented concept, 2 stars

Watch-list entry for stacklit-cli: Add a row with trigger condition "50+ GitHub stars AND practitioner write-up from production use."

No guide files modified by this evaluation.


Files Modified

  • docs/resource-evaluations/liza-mas-token-saving-cli-tools.md (this file)
  • docs/resource-evaluations/README.md (index row)
  • docs/resource-evaluations/watch-list.md (stacklit-cli row)
  • CHANGELOG.md ([Unreleased] entry)