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 instructions/vagkaratzas/token-saviour/agents-mdgit clone --depth 1 https://github.com/vagkaratzas/token-saviourWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/vagkaratzas/token-saviour/agents-md)<a href="https://agentmods.dev/instructions/vagkaratzas/token-saviour/agents-md"><img src="https://agentmods.dev/badge/instructions/vagkaratzas/token-saviour/agents-md.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.00575 | $0.00575 |
| Opus 5 | $0.00287 | $0.00287 |
| Sonnet 5 | $0.00115 | $0.00115 |
| Haiku 4.5 | $0.00057 | $0.00057 |
Grade A, and why
token-saviour AGENTS.md scanned grade A with 0 findings 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 4d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
token-saviour — spend tokens where they matter
Always-on rules for agents that read AGENTS.md (Codex CLI/app, VS Code Codex extension, and
other instruction-only hosts). Full playbook with commands and evidence:
skills/token-saviour/SKILL.md.
Token cost has four independent layers; a different tool owns each. Before acting, ask:
"which layer is this task spending tokens on?" — then use that layer's tool if it is installed.
Never pretend a tool ran; if it's missing, fall back to plain Read/Grep/Bash (installs:
skills/token-saviour/references/tool_links.md).
| When you're about to… | Use instead | Layer |
|---|---|---|
cat/read several files to find a symbol, callers, a call path, or explain a module |
serena MCP tools: get_symbols_overview, find_symbol, find_referencing_symbols (it also edits: replace_symbol_body, rename_symbol) |
code-read input |
run a noisy command — tests, build, lint, git, grep, find, big listings |
rtk proxies: rtk test <cmd>, rtk grep, rtk read, rtk find, rtk git … |
command-output input |
| write a long, chatty prose reply | caveman terse mode (keep code, commands, errors, ordered/irreversible steps verbatim) | prose output |
| generate/edit a chunk of code | Ponytail rules: YAGNI, stdlib first, no unrequested abstractions, delete over add, one runnable check | code output |
| a tiny one-file/one-line lookup | plain Read/Grep/Bash — tool overhead beats the benefit | — |
Rules of thumb (all benchmark-measured; see skills/token-saviour/references/benchmark_results.md):
- Code reads dominate the bill (~88%); switching them to serena is the highest-value move (−66% alone; the full stack measured −70%).
- One tool per layer. Never run two code-read tools; never use rtk for comprehension (~0% there); never use caveman on code or Ponytail on prose (each ~0% outside its lane).
- If a loop is dominated by test runs and Pare's MCP servers are wired in, its structured
pytestbeats rtk on that one command (−72%); rtk still owns the layer overall. - Announce what you used, one line:
🪙 token-saviour: serena + rtk + caveman(orplain tools (fallback)).
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.
- 4d ago First seen · 30 lines · 575 tokens per session scan A f17657e3a837
token-saviour AGENTS.md is an instructions file published in the GitHub repository vagkaratzas/token-saviour (10 stars, last pushed 1mo ago), licensed MIT. It adds 575 tokens to every session, about $0.0029 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
Accordion CLAUDE.md
Instructions for a-Fig/Accordion, covering claude.md — accordion, key urls, terminology, codebase map and engine — single source of truth.
tokf CLAUDE.md
Claude Code instructions for mpecan/tokf, covering tokf — development guidelines, project philosophy, commits, code quality and testing.
Accordion AGENTS.md
Instructions for a-Fig/Accordion, a project described as: 🏆 AI Hackathon 2026 @ UC Berkeley Intelligent context management for developers.
mentedb copilot-instructions.md
Instructions for nambok/mentedb, covering mentedb development instructions, project overview, workspace structure, build, test, and lint and key types.
OmniGlyph CLAUDE.md
Claude Code instructions for diegosouzapw/OmniGlyph, covering claude.md, quick start, project at a glance, key conventions and running benchmarks.
OmniGlyph GEMINI.md
Gemini CLI instructions for diegosouzapw/OmniGlyph, covering omniglyph — guidelines for gemini-based agents, non-negotiables (mirror of agents.md), file placement and validation before you claim done.