Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/edimuj/tokenlean/dedupnpx skills add edimuj/tokenlean --skill dedupgit clone --depth 1 https://github.com/edimuj/tokenleanWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00069 | $0.01642 |
| Opus 5 | $0.00034 | $0.00821 |
| Sonnet 5 | $0.00014 | $0.00328 |
| Haiku 4.5 | $0.00007 | $0.00164 |
Grade A, and why
dedup scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
import { execFileSync } from 'node:child_process'; How it starts
The opening of the file, as written. The whole thing — 164 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Dedup
tl dupes finds duplicates; removing them well is the skill. Most of the
output is noise, the wins hide in two tiers, and the mitigation that actually
sticks is non-obvious: you remove the second home, not "fix both copies."
Workflow
Scan → Triage → Mitigate → Verify → Ratchet
1. Scan
tl dupes --near # structural + near surfaced first; tests excluded by default
Via the MCP tool, near is a number (the similarity threshold), not a flag:
mcp__callmux__tl_dupes { near: 0.85 }. Passing near: true fails schema
validation — the CLI's bare --near defaults that threshold for you.
Files copied verbatim into another tool (e.g. a plugin shipped into a user's
config dir) are external contracts — their duplicate helpers are by-design,
not cleanup targets. tl dupes excludes them by default. Mark your own in
.tokenleanrc.json:
{ "externalContractFiles": ["src/my-copied-plugin.js"] }
Use tl dupes --near --include-contracts only when you deliberately want to
audit those copies too.
2. Triage by tier — signal vs. noise
tl dupes returns four tiers. Treat them differently:
- Structural + near = signal. Renamed clones and copy-paste-with-edits — the drift-prone ones. Start here.
- Exact = quick wins, but check intent first. Two trivial pass-throughs
colliding (
return [...this.x.values()]) aren't always worth a shared home. - Names = mostly noise. Idiomatic repeats (
update,close,get, logger levels). Skim, don't mine — add--no-namesonce you've glanced.
3. The "real dupe?" test
The rule that separates worth-fixing from leave-it:
- Same rule encoded twice → collapse it. Validation, response unwrapping, a path-containment check, a magic constant, a normalization step. These are the drift hazards: one copy gets a fix, the other doesn't, and a phantom bug surfaces weeks later.
- Same shape by coincidence → leave it. Logger
debug/info/warn/error, SSEemit*helpers, store thunks, thin HTTP wrappers. Merging hurts readability for zero safety gain.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 164 lines · 69 tokens per session scan A 69d5d02eaec4
dedup is a skill published in the GitHub repository edimuj/tokenlean (11 stars, last pushed 1mo ago), licensed MIT. It adds 69 tokens to every session and 1,642 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
zerodust
Sweep 100% of native gas tokens (ETH, BNB, MATIC, etc.) from EVM chains via ZeroDust, leaving exactly zero balance. Use when: user wants to exit a chain completely, consolidate dust balances, clean up wallets, or bridge remaining native tokens cross-chain. Supports 25 mainnet chains including Ethereum, Base, Arbitrum…
ctx2img
Cut context tokens by rendering it as images. ctx2img paint turns any text-shaped input (a repo, a directory, a file, markdown, stdin) into dense images that carry the full text at 60-75% fewer tokens, with stable handles and a verbatim factsheet; ctx2img read recovers guaranteed-exact text. Use before ingesting any…
r3f-animation
React Three Fiber animation - useFrame, useAnimations, spring physics, keyframes. Use when animating objects, playing GLTF animations, creating procedural motion, or implementing physics-based movement.
design-system-builder
Extract design systems from reference UI images and generate implementation-ready UI design prompts. Use when users provide UI screenshots/mockups and want to create consistent designs, generate design systems, or build MVP UIs matching reference aesthetics.
r3f-best-practices
React Three Fiber (R3F) and Poimandres ecosystem best practices. Use when writing, reviewing, or optimizing R3F code. Triggers on tasks involving @react-three/fiber, @react-three/drei, zustand, @react-three/postprocessing, @react-three/rapier, or leva.
log-error-digest
Analyze log files to troubleshoot errors, identify peak error periods, and produce error clustering, frequency statistics, and time distribution reports. Supports JSON, syslog, and Nginx formats with automatic detection. Use when a user uploads a .log file and asks to analyze errors, find patterns, debug issues, or…