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 skills add saski/arnesto --skill headroomgit clone --depth 1 https://github.com/saski/arnestoWrote 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/skills/saski/arnesto/headroom)<a href="https://agentmods.dev/skills/saski/arnesto/headroom"><img src="https://agentmods.dev/badge/skills/saski/arnesto/headroom/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/saski/arnesto/headroom"><img src="https://agentmods.dev/badge/skills/saski/arnesto/headroom.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00070 | $0.00689 |
| Opus 5.5 | $0.00028 | $0.00276 |
| Sonnet 5.5 | $0.00014 | $0.00138 |
| Haiku 4.5 | $0.00007 | $0.00069 |
Grade A, and why
headroom 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 9d 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.
How it starts
The opening of the file, as written. The whole thing — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Headroom
Treat compression as a transformation that must preserve task-critical evidence. Start with a synthetic local example, then compare representative payloads before changing any running agent's request path.
Choose the integration
- Prefer upstream query filtering, bounded reads, or RTK for simple shell output. Avoid stacking compressors without measuring the combined result.
- Use the Python library when the application owns the request payload. The
reviewed package is
headroom-ai==0.39.1, requiring Python 3.10 or newer. - A proxy,
headroom wrap, or MCP installation is a separate integration: those paths can change provider routing, credentials handling, configuration, and stored context. Inspect those changes before applying them; this skill does not authorize a client migration or imply desktop-client compatibility.
Start locally
Use the project's dependency manager, or an isolated environment for a trial.
Install the pinned base package, not [all], for the example in
local trial. Disable telemetry and update checks
before running it. The example uses no API key and makes no LLM request.
from headroom import CompressConfig, compress
result = compress(
messages,
model="gpt-4o", # Tokenizer selection for this example; no model request.
config=CompressConfig(
kompress_model="disabled",
compress_system_messages=False,
compress_user_messages=False,
protect_recent=4,
),
)
Use the actual target model's supported tokenizer for a real evaluation. A first tokenizer load can require a vocabulary download. Do not report estimated counts as measured tokens. Disabling Kompress avoids its ML-model download; it does not make initial package/tokenizer installation offline.
Check evidence preservation
- Retain the original payload separately. Library compression is lossy; full retrieval requires a working retained store and retrieval integration.
- Verify system instructions, current user intent, and recent messages remain intact. Protect tool-call IDs, relevant errors, rare outliers, and exact data needed by the task. Do not truncate away permission or policy information.
- Compare downstream answers and required assertions with and without compression. Token reduction alone is not success.
- If data is missing, counts are estimates, the store is unavailable, or correctness regresses, use the original payload and report the limitation.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 9d ago First seen · 76 lines · 70 tokens per session scan A 5a0669761414
headroom is a skill published in the GitHub repository saski/arnesto (5 stars, last pushed yesterday), licensed Unlicense. It adds 70 tokens to every session and 689 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-10-08.
Other skills, from other repositories
jetson-inference-mem-tune
Pick the serving stack and per-runtime memory flags (vLLM, SGLang, llama.cpp, TensorRT Edge-LLM) for an LLM/VLM workload on any NVIDIA Jetson.
kg-builder
Designs and builds knowledge graphs from documents — ontology modeling with domain/range constraints, entity/relation/event extraction, entity resolution, provenance and supersession, and GraphRAG serving. Use when asked to "build a knowledge graph", "design an ontology", "extract entities and relations", "deduplicate…
rag-auditor
Evaluates RAG pipeline quality across retrieval (precision, recall, MRR) and generation (groundedness, hallucination rate). Triggers on: "audit RAG pipeline", "RAG quality", "hallucination detection", "why is RAG failing", "grounding check". NOT for general architecture audits, use architecture-reviewer.
rag-implementation
Comprehensive guide to implementing RAG systems including vector database selection, chunking strategies, embedding models, and retrieval optimization. Use when building RAG systems, implementing semantic search, optimizing retrieval quality, or debugging RAG performance issues.
rag-architect
Use when the user asks to design a RAG pipeline, choose a chunking strategy or embedding model, pick a vector database, or evaluate retrieval quality (precision@k, recall@k, NDCG). Examples: 'design a RAG system for our docs', 'what chunk size should I use for this corpus', 'evaluate my retriever against ground…
chroma-hybrid-search
Local hybrid retrieval (BM25 + ChromaDB vector + BGE-Reranker) over /.deep-memory hot and cold stores. Use when high-accuracy code/solution retrieval is needed and hallucination must be minimized. Typically invoked by deep-memory.