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 houshuang/limbic --skill thin-worker-briefgit clone --depth 1 https://github.com/houshuang/limbicWrote 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/houshuang/limbic/thin-worker-brief)<a href="https://agentmods.dev/skills/houshuang/limbic/thin-worker-brief"><img src="https://agentmods.dev/badge/skills/houshuang/limbic/thin-worker-brief/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/houshuang/limbic/thin-worker-brief"><img src="https://agentmods.dev/badge/skills/houshuang/limbic/thin-worker-brief.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 | $0.00103 | $0.01108 |
| Opus 5.5 | $0.00041 | $0.00443 |
| Sonnet 5.5 | $0.00021 | $0.00222 |
| Haiku 4.5 | $0.00010 | $0.00111 |
Grade A, and why
thin-worker-brief 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 — 92 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Thin worker brief
Every subagent re-reads its whole entrance context on each turn. Measured
median entrance fee: 51.5K tokens, before the worker does anything. A
50K-token instruction stack is therefore not documentation, it is a per-turn
tax — and in the project that had one, workers still made 1,516 hand-rolled
sqlite3 calls and rediscovered the schema 332 times in a month.
So: decide whether to delegate at all, then make the brief small.
Delegate only when the work is not a call
A subagent is the right tool for judgement, search across an unknown surface,
and review. It is the wrong tool for applying a codebook to N documents — that
is packet-worker, and the gap is 95×. Before spawning, ask whether a
stateless call with a fixed schema does the job.
The brief: ≤ 2K tokens
A WORKER.md of four parts, and nothing else:
- Schema card. The tables or record shape the worker touches, with the
traps inline ("
normalized_nameis onlylower(name): no diacritic folding"; "prf_idis TEXT"; "a title can be numeric-looking, never assume a string"). One screen. This replaces the rediscovery loop. - The lookup CLI, with four example invocations. Not "query the
database" — the exact command. A documented-but-broken CLI is worse than
none: one crashed on an integer title, took 5 s and missed the obvious
query, and was called twice in a month against 1,516 raw
sqlite3calls. Verify yours runs before you ship the brief. - The proposal template, and the statement that the worker proposes and never writes.
- Fetch helpers and flags the worker will otherwise reinvent (the no-metrics query param, the offline snapshot path, the rate limit).
A named reading list, never "read the docs". Name the file and the
section: CODEBOOK.md §11 borderline rules, not CODEBOOK.md. An unbounded
pointer is read in full, on every turn, by every worker.
State the pinned revision (a SHA, not "main") and the worker's write scope as literal paths.
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 · 92 lines · 103 tokens per session scan A 0f3ba56747e8
thin-worker-brief is a skill published in the GitHub repository houshuang/limbic (3 stars, last pushed 3d ago), licensed MIT. It adds 103 tokens to every session and 1,108 once invoked, about $0.0004 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 2026-09-22.
Other skills, from other repositories
sentence-transformers
Framework for state-of-the-art sentence, text, and image embeddings. Provides 5000+ pre-trained models for semantic similarity, clustering, and retrieval. Supports multilingual, domain-specific, and multimodal models. Use for generating embeddings for RAG, semantic search, or similarity tasks. Best for production…
cortex-distill
A skill for turning raw session records into organised notes and project documents. It uses a personal vault, which is a folder for storing and retrieving knowledge.
cortex-broadcast
Propagate insights from a distilled Raw into related existing Notes/Projects pages through conversational per-page edits.
cortex-query
A search tool for finding and retrieving information from Cortex, an external personal memory made up of stored notes and project pages.
cortex-evolve
A tool that saves technical notes and project decisions to a Cortex vault, which is an Obsidian note collection stored as files.
sentence-transformers
Framework for state-of-the-art sentence, text, and image embeddings. Provides 5000+ pre-trained models for semantic similarity, clustering, and retrieval. Supports multilingual, domain-specific, and multimodal models. Use for generating embeddings for RAG, semantic search, or similarity tasks. Best for production…