GitHub - vagkaratzas/token-saviour: Agent skill that routes each coding task to the most token-efficient tool per layer: serena for code reads, rtk for command output, caveman for prose, Ponytail for generated code — −70% tokens measured in a 9-tool benchmark. Installable as a Claude Code plugin or Codex skill. · GitHub

/" data-turbo-transient="true" />

Skip to content

Search or jump to...

Search code, repositories, users, issues, pull requests...

-->

Search

Clear

Search syntax tips

Provide feedback

--><br>We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

-->

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

Sign in

/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up

Appearance settings

Resetting focus

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

vagkaratzas

token-saviour

Public

Notifications<br>You must be signed in to change notification settings

Fork

Star

main

BranchesTags

Go to file

CodeOpen more actions menu

Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit

History<br>4 Commits<br>4 Commits

.claude-plugin

.claude-plugin

.codex-plugin

.codex-plugin

assets

assets

skills/token-saviour

skills/token-saviour

AGENTS.md

AGENTS.md

LICENSE

LICENSE

README.md

README.md

View all files

Repository files navigation

token-saviour

Spend tokens where they matter. An agent skill that routes every coding task to the most<br>token-efficient tool for the layer it stresses — instead of reflexively reading whole files,<br>dumping raw command output into context, or writing more code than asked.

Token cost has four independent layers , and a different tool owns each. token-saviour picks<br>one winner per layer — the best measured combination from a<br>9-tool benchmark on a realistic<br>Python codebase:

Layer<br>Tool<br>Measured

Code-read input (symbols, callers, call paths, architecture)<br>serena<br>−66% alone — the dominant cost

Command-output input (tests, builds, git, grep, listings)<br>rtk<br>−65% on a test run, −36% on command tasks

Generated prose output (chatty replies, write-ups)<br>caveman<br>−6% on prose answers (grows with chattiness)

Generated code output (implementations you write)<br>Ponytail<br>~−40% on a verbose implementation (illustrative)

Stacked (serena + rtk + caveman): −69.6% total tokens on the comprehension suite. For<br>code-generation work, swap caveman → Ponytail (≈ −64%).

The picks were re-validated (2026-07) against newer entrants — archex, Pare, lazy-cat — and every<br>incumbent held its layer. The runner-ups, the niches where they flip (e.g. Pare's structured<br>pytest for test-dominated loops), and the full evidence live in<br>skills/token-saviour/references/.

Install

Claude Code (plugin)

/plugin marketplace add vagkaratzas/token-saviour<br>/plugin install token-saviour@token-saviour

Or as a plain personal skill, no plugin system involved:

git clone https://github.com/vagkaratzas/token-saviour<br>mkdir -p ~/.claude/skills<br>cp -r token-saviour/skills/token-saviour ~/.claude/skills/

Codex

codex plugin marketplace add vagkaratzas/token-saviour<br>codex plugin add token-saviour@token-saviour

In Codex the skill is invoked with @token-saviour. The VS Code Codex extension and the Codex<br>app read AGENTS.md, which this repo ships — so it also works from the repo root with no<br>setup, and cp AGENTS.md ~/.codex/AGENTS.md makes the always-on rules global.

Any other agent

Copy AGENTS.md (compact, always-on ruleset) or skills/token-saviour/SKILL.md (full<br>playbook) into whatever instruction file your agent reads.

What the skill does

Classifies the task by the layer it stresses: reading code, reading command output,<br>writing prose, or writing code.

Routes to that layer's tool with concrete commands (serena MCP calls, rtk proxies,<br>caveman terse mode, Ponytail rules) — and degrades gracefully to plain Read/Grep/Bash when a<br>tool isn't installed (install/verify commands in<br>references/tool_links.md).

Refuses anti-patterns : two code-read tools at once, rtk for comprehension, caveman on<br>code, Ponytail on prose, tooling-up trivial one-line lookups.

Announces what it used : 🪙 token-saviour: serena + rtk + caveman.

The tools it routes to (install the smallest set you need)

serena — uv tool install -p 3.13 serena-agent — the one code-read tool (LSP symbols;<br>also does semantic edits/renames).

rtk — brew install rtk or cargo install --git https://github.com/rtk-ai/rtk — add<br>only for noisy command loops.

caveman — see repo — add only when prose<br>brevity is the bottleneck.

Ponytail — /plugin marketplace add DietrichGebert/ponytail — add only when<br>code-generation work is the...