内容
<div align="center">
**English** · [**中文**](README.zh-CN.md)
<img src=".github/assets/banner.svg" alt="Agent-Body — plugins as organs: 26 curated organs across 8 systems, one heartbeat, 84.7% of tool-schema tokens gated away, 0 chronic wounds" width="92%">
# Agent‑Body
**Load the tools your task needs. Give your agent memory, reflexes, and a recovery loop.**
A plugin layer for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) that routes intent to capabilities and keeps unused tool definitions out of the model’s context.
**26 organ identities · 84.7% of tool-schema tokens gated cold-start · 200+ offline assertions · MIT**
</div>
Your agent may have hundreds of tools. A single task usually needs only a few. Agent-Body adds a runtime layer that decides which capabilities to expose, remembers outcomes, and attributes failures before retrying.
**[Watch the recorded demo](https://1420079678-ctrl.github.io/agent-body/en.html)** · **[Read the benchmark](benchmarks/results/REPORT.md)** · **[Install in DeepSeek Harness](#install-in-one-line)** · [中文](README.zh-CN.md)
| Problem | What Agent-Body does | Evidence you can inspect |
| --- | --- | --- |
| Tool definitions consume context on every turn | Gates schemas by task intent | 48-command offline benchmark: **55,154 → 8,433 estimated schema tokens on average** |
| Routing adds another model decision | Routes commands through deterministic intent rules | Offline end-to-end demo; routing needs no model call |
| Failures trigger unhelpful retries | Classifies causes and selects a remedy; argument errors are not retried | Recovery trace in the recorded runtime output |
| Useful outcomes disappear between sessions | Adds persistent memory and idle consolidation through the Cortex organ | [Architecture](ARCHITECTURE.md) and included plugin source |
The **84.71% reduction is a cold-start tool-schema estimate**, not a total-prompt or billing claim. The benchmark exposes **86.67% of task-declared required capabilities on the first turn**; six are capped and must be retrieved via `body_call`. A recorded installation with history shows **58% schema gating**. [See the full trade-off and methodology](benchmarks/results/REPORT.md).
The body metaphor maps to concrete components: plugins are organs, intent routing is the nerve bus, deterministic event handlers are reflexes, and the heartbeat coordinates state. **This is an early-stage, Windows-first project.** Host integration requires DeepSeek Harness; the offline core demo below runs independently.
**Try it in 30 seconds — no install, no host, no API key.** The core imports nothing outside Node built-ins, so a fresh
clone runs the real end-to-end chain offline:
```bash
git clone https://github.com/1420079678-ctrl/agent-body && cd agent-body
npm run demo # command → impulse → dispatch → execute → attribute → reflex fires
npm run check # the gate CI runs: constant tables, catalog, tests, benchmark — all offline
```
Here is that same body while it runs. Not a mockup — the frames are real tool output from a real install, rendered by
`tools/make-demo-gif.py` out of [`docs/demo/captured.json`](docs/demo/captured.json), which records the raw output. The
renderer **refuses to draw a number that is not in that file**:
<a href="docs/demo/captured.json"><img src="docs/demo/agent-body.gif" alt="Live output: 43 organs, 332/332 capabilities claimed, 20/20 reflexes armed, 82,126 to 33,348 tool-schema tokens, 98% heal rate, and one Chinese command that innervates 4 organs with zero model calls" width="100%"></a>
Read it left to right: anatomy and vitals, then the token ledger, then a **Chinese command becoming a nerve impulse**
that innervates four organs — each told *which* capability to fire, with zero model calls spent deciding the route —
and finally a wound attributed to `arg_error`, prescribed **do-not-retry**, and closed by re-check.
**Already running DeepSeek Harness?** One command installs the body kernel, the memory organ and the context engine:
```powershell
$rel = "https://cdn.jsdelivr.net/gh/1420079678-ctrl/agent-body@v0.1.2/dist"
dsh plugin --profile web add "$rel/dsh-external-dsh-organism-0.1.1.tgz" "$rel/dsh-external-dsh-cortex-0.1.1.tgz" "$rel/dsh-external-dsh-zero-residence-0.1.0.tgz"
```
Restart the harness and `body_status` lists the organs. **The claim you can check for yourself:** the tool-schema block
of the prompt drops **84.7%** on a cold start across 48 representative commands, and `npm run bench:check` fails the
build if that number drifts. The scope is stated wherever the number appears — tool-schema tokens only, not the whole
prompt. On a body that has been running for a while it is **58%** (127 of 332 visible); that figure, and why it
declines, are in [Token economy](#token-economy).
⭐ **[Star the repository](https://github.com/1420079678-ctrl/agent-body/stargazers)** if you want it to keep tracking
the host closely — it is a one-person project. A star helps other developers discover it; a reproducible issue or a new organ helps improve it.
---
<div align="center">
[](#organ-catalog)
[](#organ-catalog)
[](#token-economy)
[](benchmarks/results/REPORT.md)
[](#verify-it-yourself)
[](#quick-start)
[](LICENSE)
[](https://github.com/1420079678-ctrl/agent-body/stargazers)
[](https://github.com/1420079678-ctrl/agent-body/issues)
[](https://dshfind.com/en/plugins/1420079678-ctrl/agent-body)
[Architecture](ARCHITECTURE.md) · [Organ Catalog](catalog/organs.json) · [Benchmark](benchmarks/results/REPORT.md) · [Roadmap](ROADMAP.md) · [**中文文档**](README.zh-CN.md)
[Official DSH discussion — **Show Your Plugins!**](https://github.com/deepseek-ai/deepseek-harness/discussions/7555) · the channel the harness `CONTRIBUTING` points plugin authors to
**v0.1.1** · MIT · Windows-first (Node 22.19 / 24) · **26 organ identities in the catalog, realised by 24 plugin packages in this repository** · [release notes](https://github.com/1420079678-ctrl/agent-body/releases/tag/v0.1.1)
[](https://1420079678-ctrl.github.io/agent-body/en.html)
▶ **[Open the recorded demo](https://1420079678-ctrl.github.io/agent-body/en.html)** ([中文](https://1420079678-ctrl.github.io/agent-body/)) — a recorded replay of a real install: the heartbeat, the organs, the pulse stream, the healing ledger and the token gate, with nothing installed. It is generated from the runtime files (`vitals.json`, `bloodstream.json`, `pulse.jsonl`), not retyped from screenshots.
**Contents** · [Where it's listed](#where-its-listed) · [Install in one line](#install-in-one-line) · [Why this exists](#why-this-exists) · [What makes it different](#what-makes-it-different) · [The five biological layers](#the-five-biological-layers) · [Architecture at a glance](#architecture-at-a-glance) · [Organ catalog](#organ-catalog) · [Quick start](#quick-start) · [Write your own organ](#write-your-own-organ) · [Verify it yourself](#verify-it-yourself) · [When not to use this](#when-not-to-use-this) · [FAQ](#faq) · [Repository layout](#repository-layout) · [Contributing](#contributing) · [Roadmap](#roadmap)
</div>
---
## Where it's listed
Checked, not claimed — every entry below resolves today:
| Listing | What it is |
| --- | --- |
| [`awesome-dsh-plugin`](https://github.com/awesome-dsh-plugin/awesome-dsh-plugin) · 16.7k ★ | the main DSH plugin directory; both `dsh-organism` and `dsh-cortex` are indexed as separate entries |
| [dshfind](https://dshfind.com/en/plugins/1420079678-ctrl/agent-body) | DSH plugin search engine, with a per-repository page |
| [`imsai-sh/awesome-deepseek-harness-plugins`](https://github.com/imsai-sh/awesome-deepseek-harness-plugins) | plugin hub with machine-readable catalogue entries |
| [`bruc3van/awesome-dsh-plugin`](https://github.com/bruc3van/awesome-dsh-plugin) | daily-crawled DSH plugin list, human-reviewed |
| [`billLiao/awesome-dsh-plugin`](https://github.com/billLiao/awesome-dsh-plugin) | curated list, memory category |
| [`unStone/dsh-xray`](https://github.com/unStone/dsh-xray) | declared-capabilities scanner; carries a page for this repository |
| [`linny006/agent-framework-radar`](https://github.com/linny006/agent-framework-radar) · [`llmops-radar`](https://github.com/linny006/llmops-radar) | live indexes of newly shipping agent frameworks and LLMOps tooling |
It is also the subject of the harness's official showcase thread —
[**Show Your Plugins!** discussion #7555](https://github.com/deepseek-ai/deepseek-harness/discussions/7555).
---
## Install in one line
Two supported paths — start with the first, switch to the second if your network blocks `github.com`.
### Path 1 — GitHub release assets
Versioned assets built by CI. `releases/latest/download/` always resolves to the newest release, so this command does
not need editing when a version is bumped.
```powershell
$rel = "https://github.com/1420079678-ctrl/agent-body/releases/latest/download"
dsh plugin --profile web add `
"$rel/dsh-external-dsh-organism-0.1.1.tgz" `
"$rel/dsh-external-dsh-cortex-0.1.1.tgz" `
"$rel/dsh-external-dsh-zero-residence-0.1.0.tgz"
```
```bash
rel=https://github.com/1420079678-ctrl/agent-body/releases/latest/download
dsh plugin --profile web add \
"$rel/dsh-external-dsh-organism-0.1.1.tgz" \
"$rel/dsh-external-dsh-cortex-0.1.1.tgz" \
"$rel/dsh-external-dsh-zero-residence-0.1.0.tgz"
```
### Path 2 — served from this repository over jsDelivr
The same tarballs, committed under [`dist/`](dist/) and served by a CDN with mainland nodes. Reach for this when Path 1
hangs: those URLs 302-redirect to `objects.githubusercontent.com`, a hop that is blocked on some networks. The symptom
is `fetch failed` with `downloaded 0` **while dependency resolution succeeds** — the download hop is the problem, not
the packages.
```powershell
$rel = "https://cdn.jsdelivr.net/gh/1420079678-ctrl/agent-body@v0.1.2/dist"
dsh plugin --profile web add `
"$rel/dsh-external-dsh-organism-0.1.1.tgz" `
"$rel/dsh-external-dsh-cortex-0.1.1.tgz" `
"$rel/dsh-external-dsh-zero-residence-0.1.0.tgz"
```
```bash
rel=https://cdn.jsdelivr.net/gh/1420079678-ctrl/agent-body@v0.1.2/dist
dsh plugin --profile web add \
"$rel/dsh-external-dsh-organism-0.1.1.tgz" \
"$rel/dsh-external-dsh-cortex-0.1.1.tgz" \
"$rel/dsh-external-dsh-zero-residence-0.1.0.tgz"
```
`dsh plugin` forwards its arguments to `pnpm add` inside the profile, so several packages install in one command.
Restart the harness afterwards; `body_status` should list the organs. The other organs in the catalog are one tarball
each — see [Quick start](#quick-start).
<details>
<summary>If neither path works — install fully offline</summary>
Every tarball built by CI is committed under [`dist/`](dist/), and
[`dist/agent-body-offline-kit.zip`](dist/agent-body-offline-kit.zip) bundles all ten plus a one-command installer.
Download it from any machine that can reach this repository, copy it across, then:
```powershell
Expand-Archive .\agent-body-offline-kit.zip -DestinationPath .
pwsh .\install-agent-body.ps1
```
</details>
**Why not `npm install` yet.** These packages are **not published to the npm registry** — the names above return 404
from `registry.npmjs.org` today, and nothing here pretends otherwise. That is the next step and it is tracked as an
open issue: once they are on npm, `dsh plugin add @dsh-external/dsh-organism` becomes a single short command. Until
then, the paths above are the supported ones.
## Start with the evidence
| If you want… | Open… | What you can check |
| --- | --- | --- |
| to see it run without installing anything | `npm run demo` | a real command → impulse → dispatch → execute → attribute → reflex chain, offline, no key |
| fewer tool schemas in your prompt | [`benchmarks/results/REPORT.md`](benchmarks/results/REPORT.md) · `npm run bench:check` | **84.71%** of tool-schema tokens gated away on 48 commands (cold-start, reproducible); 58% with history (2026-09-25) |
| failures to stop repeating | `body_heal` | attribution before retry: `tool_missing` / `arg_error` / `permission` / `timeout` / `network` / `not_found` / `conflict`. Measured on a development install: **198 healed, 0 open, 99% heal rate** |
| to know the system is still alive | `body_status` · `body_heart` | 43 organs, **332/332** capabilities claimed, beat #1121, architecture integrity 6/6 *(development install)* |
| to lose an organ without losing the body | `body_organ action=integrity` | the six core pieces depend on no single organ; `body_call` compensates from the organ that overlaps most |
| to write your own organ | [`docs/ORGAN_SDK.md`](docs/ORGAN_SDK.md) | `defineOrgan()` validates at declaration and throws with a field path; `npm run check` gates the whole repository |
---
## Why this exists
Every agent framework eventually becomes the same thing: a pile of tools, a growing list in the system prompt, and an
agent that forgets everything between sessions.
It re-reads the same failure twice before learning it exists. It cannot tell you which of its own capabilities are
broken. It has no idea that a plugin it depends on went offline — it just calls it and fails. And you keep paying tokens
for tool schemas nobody calls.
**Agent‑Body takes the opposite bet: treat the plugin system as an organism.**
Every plugin declares itself an *organ* — with capabilities, senses, and reflexes. Everything else is physiology: a
nervous system that routes your commands to the right organ, a heart that pumps state through the whole body on a
rhythm, reflex arcs that fire without a single model call, and a healing loop that attributes every failure before it
even thinks about retrying.
The result is a system that gets **measurably better the more you use it**, and that stays standing when you rip parts
of it out.
---
## What makes it different
### 🫀 It has a heartbeat, not a loop
A real pacemaker runs at a 1s base clock with a **variable interval** — 3× faster on critical alerts, 2× on fatigue,
2× slower when the body has been silent for a long while. Every beat packages the current operator directives, the
body's proprioception and the homeostasis warnings into a blood packet, writes it to `bloodstream.json`, and broadcasts
it as `organism/heartbeat`. Any organ can hook the circulation with one line:
```ts
ctx.on('organism/heartbeat', (blood) => { /* your organ now has a pulse */ })
```
Circulation is a loop, not a firehose: organs return blood through `organism/venous`, and newly learned knowledge is
oxygenated (`organism/oxygenated`) before it is pumped system‑wide.
### 🧠 Your sentence becomes a nerve impulse
You type a command. Before the model even reasons about it, `agent/pre-step` converts it into an impulse, deterministically
*innervates* the organs that should handle it, and broadcasts on `organism/impulse` — telling each organ **which of its
capabilities to use**.
Then the body learns the route. When an innervated organ actually delivers, a Hebbian synapse strengthens
"this kind of command → that organ"; when it fails, the weight decays. Dispatch order reorders itself from experience,
persisted in `synapses.json`.
```
you: "抓取这个站点并抽取结构化数据"
impulse → sensory(web-crawl) · executive(plan) · immune(verify)
→ each organ told which capability to fire
```
### 🩹 Self‑healing is a closed loop, not a slogan
```
failure → deterministic attribution → prescription → RE‑CHECK
```
Attribution is not guessing: `tool_missing` / `arg_error` / `permission` / `timeout` / `network` / `not_found` /
`conflict` / `unknown`. The prescription table decides the response — read‑only remedies execute automatically,
side‑effecting ones wait for the brain to rule, and **`arg_error` is never auto‑retried** (retrying a wrong argument
just amplifies the mistake).
The wound only closes when **that organ next succeeds** — the system does not declare itself healed. Anything still open
after 5 minutes is marked `chronic` and stops spinning.
> Measured on the development install: **21 wounds healed · 100% heal rate · 0 chronic.** An `arg_error` wound closed
> 3.5s after the underlying fix landed.
### 🌱 It trains itself — and forgets on purpose
Three learning channels run continuously, none of them requiring you to teach anything:
1. **Synaptic learning** — success reinforces "command class → organ", failure weakens it.
2. **Reflex self‑authoring** — the same tool × cause failing 3 times makes the system **write its own reflex arc**
(`R-auto-<cause>-<tool>`), which then fires automatically on the next occurrence.
3. **Chain crystallization** — any cross‑organ chain that completes with ≥3 steps is fixed into a replayable skill.
Forgetting is equally deliberate: synapses decay on a 30‑minute half‑life and get pruned below `|w| < 0.2`; reflexes
that fired ≥5 times with zero contribution are retired; skills that keep failing are forgotten.
`body_heal action=rehab` restores an organ to its healthy baseline — **damage cleared, wisdom kept**: fatigue and wounds
reset, learned synapses, reflexes and skills untouched.
### 💤 It sleeps, consolidates, and remembers
Idle for 2 minutes → light sleep. 5 minutes → deep sleep, and every 2 minutes of deep sleep runs a **consolidation
pass with zero model calls** — pure deterministic rules mining the experiences of the day into long‑term memory:
| Card | Mined from |
| --- | --- |
| `pitfall` | same tool failing ≥3 times in a row |
| `playbook` | a cross‑tool success chain repeating ≥2 times (trivial same‑tool repeats filtered out) |
| `hotspot` | ≥8 calls with <70% success rate |
| `unresolved` | ≥3 failures accumulated after the last success |
| `fact` | explicit operator facts |
New commands trigger deterministic recall — tags > title > body, multiplied by weight — and the most relevant cards are
injected into context. Unused cards decay on a 168‑hour half‑life and **archive instead of being deleted**.
### 🪶 Zero‑residence context
Instead of lossy LLM summarization, the engine replaces resident content with **deterministic pointers** — and every
masked byte stays recoverable verbatim from the session log:
| Tool | Purpose |
| --- | --- |
| `zr_compact` | arm a forced compaction that bypasses the ratio threshold |
| `zr_recall` | rebuild masked content verbatim, by tool call id or session seq |
| `zr_ledger` | quantify the attention integral, the three‑segment cost split, and the compression ratio |
| `zr_fast` | run long commands asynchronously and never block the turn |
### 🧩 Organs are optional — the architecture is not
Six kernels depend on no single organ: **nerve bus · heart pump · directive layer · reflex engine · dissector ·
impulse conduction**.
Unplug a plugin and the body keeps running. Uninstalling `dsh-office-docs` was detected immediately — the `craft` organ
went offline and its capability was **compensated by the closest overlapping organ**, core integrity still green. An
organ that is not installed at all is reported as *not installed*, not as *broken* — which eliminated a 15‑second
warning storm from a phantom organ.
<a id="token-economy"></a>
### 📉 Token economy as a first‑class concern
Tool schemas are shown on demand, gated by what the current turn is actually about.
**Scope of every number below: the tool‑schema block of the prompt only** — the `name` + `description` +
JSON‑schema of every tool definition. Not the system prompt, not conversation history, not tool results.
> **84.7% of tool‑schema tokens gated away (cold-start scope)** — `55,154` → `8,433` on average across 48 representative commands
> (median 85.7%, worst case 75.2%), out of **256 capability definitions**. Everything else stays one `body_call` away.
That number is produced by the benchmark in this repository and is **reproducible on your machine**:
```bash
npm run bench # regenerate benchmarks/results/REPORT.md
npm run bench:check # exit non‑zero if it drifts from the committed baseline
```
Two honest caveats, because the headline is easy to over‑read:
- **Cold‑start scope.** The figure above assumes the body has no recent activity — only the current command decides
what is revealed. On a body with real run history, recently‑used and high‑trust organs stay hot, the visible set
grows, and savings drop. Measured live figures, newest first:
| Measured | Basis | Visible | Tool-schema tokens saved |
| --- | --- | --- | --- |
| 2026‑09‑25 | this install (332 capabilities, real run history) | 127 / 332 | **58%** |
| 2026‑09‑11 | snapshot committed at `benchmarks/corpus/trace-live-gate.json` | 64 / 256 | 74% |
| earlier | live snapshot quoted by historical README revisions | — | 82% |
Both extremes are real; **always quote the scope with the number**. The cold‑start figure is the reproducible one —
run it yourself. The live figure is what you will actually see on a body that has been running for a while, and it
declines as the body grows, because more history keeps more organs hot. That worsening number is published here on
purpose: you would find it the moment you ran `body_tokens` yourself.
- **10 of 48 commands need a second hop.** A per‑organ cap of 10 capabilities means large organs (the attack organ
has 49) get truncated, and a few intents do not route to the organ that owns the capability. Those resolve through
`body_call`, but they are **not** free. The benchmark classifies every miss as *bug* / *capped* / *unrouted* and
lists them individually in [`benchmarks/results/REPORT.md`](benchmarks/results/REPORT.md).
---
## The five biological layers
| Layer | What it is | Count |
| --- | --- | --- |
| **Individual** | this body (the running install) | 1 |
| **System** | eight body systems: executive / nervous / immune / sensory / motor / memory / metabolic / endocrine | 8 |
| **Organ** | a plugin, obeying one contract: sense → reflex → effect → homeostasis | **26 organ identities** in [`catalog/organs.json`](catalog/organs.json), realised by **24** installable packages (23 in the table below, plus one legacy package kept for rollback) |
| **Tissue** | functional clustering inside an organ: sensing / inspection / effect / synthesis / memory / regulation / clearance / metering / matrix | 9 classes |
| **Cell** | a single capability unit (one tool) | counted at runtime |
Every layer is observable: `body_map` (organs + systems), `body_cell` (cells + tissues), `body_status` (vitals).
Undeclared plugins are auto‑promoted to *autonomic organs* — on the development install, **332/332 capabilities were
claimed, zero orphans**. (332 is what that live install reports through `body_status`; the frozen benchmark corpus
below is a separate, committed set of 256 capability definitions — the two numbers measure different things and are
labelled wherever they appear.)
---
## Architecture at a glance
```mermaid
graph TD
OP([Operator command]) -->|nerve impulse| NERVE["body_nerve · innervation"]
NERVE --> BUS{{"organism/impulse"}}
LAW(["Operator directives"]) --> HEART
HEART[["Heart · organism/heartbeat"]] --> BUS
BUS --> ORGANS["organs · capabilities"]
ORGANS -->|"tools/result"| REFLEX["Reflex arcs · zero tokens"]
ORGANS -->|failure| HEAL["Self-healing · attribute → prescribe → re-check"]
ORGANS -->|experience| CORTEX["Cortex · sleep / consolidate / memory"]
CORTEX -->|"recall"| ORGANS
HEAL --> VITALS["Vitals · vitals.json"]
VITALS --> HEART
ORGANS -->|venous| HEART
```
Full detail — three kernels, five life mechanisms, event contracts, on‑disk state, verification methodology — lives in
**[ARCHITECTURE.md](ARCHITECTURE.md)**.
---
## Organ catalog
Every entry below is a real plugin under `workspace/plugins/`. Five core organs ship a full **offline regression suite**
(see [Verify it yourself](#verify-it-yourself)); the remaining organs are exercised through their replay scripts.
| Organ | Plugin | System | Signature capabilities |
| --- | --- | --- | --- |
| **Organism kernel** | `dsh-organism` | nervous · endocrine | `body_map` `body_cell` `body_status` `body_heart` `body_law` `body_nerve` `body_call` `body_reflex` `body_heal` `body_skill` `body_organ` `body_pulse` `body_tokens` |
| **Cortex** | `dsh-cortex` | memory | `cortex_sleep` `cortex_memory` `cortex_homeo` |
| **Zero‑residence** | `dsh-zero-residence` | metabolic | `zr_compact` `zr_recall` `zr_ledger` `zr_fast` |
| **Web crawl** | `dsh-web-crawl` | sensory | `webcrawl` `webcrawl_site` `webcrawl_map` `webcrawl_extract` `webcrawl_doc` `webcrawl_http` `webcrawl_links` `webcrawl_status` |
| **Browser ultimate** | `dsh-browser-ultimate` | sensory | real browser engine (logged‑in sessions, anti‑bot resilience, CDP) + readability‑style extraction |
| **War bridge** | `dsh-war-bridge` | nervous | `ida` (IDA Pro MCP bridge) `war_status` `war_case` `war_memory` |
| **PentAGI** | `dsh-pentagi` | executive | `agi_plan` `agi_next` `agi_flow` `agi_memory` `agi_reflect` `agi_report` `agi_team` |
| **Sec workbench** | `dsh-sec-workbench` | immune | 49 `sec_*` capabilities: `sec_route` `sec_scope` `sec_evidence` `sec_webscan` `sec_webtest` `sec_exec` `sec_jwt` `sec_hashoff` `sec_brute` `sec_lateral` `sec_stealth` `sec_journal` `sec_report` … |
| **Reverse skill** | `dsh-reverse-skill` | immune | `rev_route` (44 routing rules) `rev_case` `rev_doctrine` `rev_playbook` `rev_journal` `rev_toolindex` |
| **Vuln remediator** | `dsh-vuln-remediator` | immune | `vuln_scan` `vuln_cve` `vuln_sbom` `vuln_priority` `vuln_patch_gen` `vuln_patch_verify` `vuln_remediate` `vuln_plan` `vuln_knowledge` |
| **Vuln daily loop** | `dsh-vuln-mastery-loop` | immune | scheduled daily practice loop with reporting |
| **Quant OS** | `dsh-quant` | executive | 59 `quant_*` capabilities: data / alpha / ML / risk / execution, `quant_research_pipeline` `quant_backtest` `quant_walk_forward` `quant_portfolio_optimize` `quant_stress_test` … |
| **Mastery loop** | `dsh-mastery-loop` | endocrine | `study_orient` `study_deconstruct` `study_model` `study_diagnose` `study_transfer` `study_review` `study_path` |
| **Academic research** | `dsh-academic-research` | executive | `ars_pipeline` (10‑stage state machine) `ars_review` (5‑seat panel) `ars_paper_plan` `ars_integrity` `ars_metrics` |
| **Office docs** | `dsh-office-docs` | motor | PDF / DOCX / PPTX / XLSX build, extract, render‑verify, pandoc convert |
| **Anatomy panel** | `dsh-anatomy-panel` | nervous | live vitals at `/anatomy`: heartbeat, wounds, learning counters, organ call counts and the real pulse stream - read from the organism runtime files on every request |
| **Social card** | `dsh-social-card` | motor | `social_card_scaffold` `social_card_render` `social_card_validate` `social_card_docs` |
| **Agent teams** | `dsh-agent-teams-pro` | executive | captain + members, task dependencies, messaging, live web panel |
| **MiroFish** | `dsh-mirofish` | executive | `mirofish_status` `mirofish_api` — swarm‑intelligence prediction engine client |
| **Minimal gray** | `dsh-minimal-gray` | metabolic | minimal agent preset, platform‑adaptive shell |
| **DingTalk bridge** | `dsh-dingtalk-bridge` | nervous | Stream long‑connection bot bridge into the harness |
| **Fund scan** | `dsh-daily-fund-scan` | executive | scheduled fund pool scan + advice report |
| **File chips** | `dsh-file-chips` | sensory | attachment chips in the composer |
| *(legacy)* | `dsh-crawl4ai` | sensory | kept for rollback; superseded by `dsh-web-crawl` |
---
## Quick start
### See it work first — no install, no host, no API key
```bash
git clone https://github.com/1420079678-ctrl/agent-body && cd agent-body
npm run demo # command → impulse → dispatch → execute → attribute → reflex fires
npm run check # constant tables + catalog + message sources + 30 tests + benchmark, all offline
```
There is nothing to install. The core packages import nothing outside Node built‑ins, so the demo runs a real
end‑to‑end chain against the committed 256‑capability corpus on a fresh clone.
Then read [`benchmarks/results/REPORT.md`](benchmarks/results/REPORT.md) to see how the token claim was measured, and
[`ROADMAP.md`](ROADMAP.md) to see what is deliberately *not* being built next.
### Install an organ in one command
The self‑contained organs ship a **prebuilt tarball** on the release page — no clone, no build step. The package declares
`dsh.bundle`, so installing it also appends it to `dsh.profile.bundles` and the next start mounts it:
```powershell
dsh plugin --profile web add https://github.com/1420079678-ctrl/agent-body/releases/download/v0.1.1/dsh-external-dsh-organism-0.1.1.tgz
```
Ten organs ship this way — the ones that need no external toolchain. Append the tarball name to the release URL:
| organ | tarball | what it adds |
| --- | --- | --- |
| `dsh-organism` | `dsh-external-dsh-organism-0.1.1.tgz` | the body kernel: anatomy, vitals, heartbeat pump, nerve impulses, reflex arcs, self-healing ledger, on-demand schema gating |
| `dsh-cortex` | `dsh-external-dsh-cortex-0.1.1.tgz` | sleep phases, deterministic consolidation, long-term memory, alert de-noising |
| `dsh-zero-residence` | `dsh-external-dsh-zero-residence-0.1.0.tgz` | zero-residence context: evict, keep a pointer, rebuild the payload verbatim on demand |
| `dsh-mastery-loop` | `dsh-external-dsh-mastery-loop-0.0.1.tgz` | subject-agnostic mastery tutor: orient → deconstruct → model → diagnose → transfer → review |
| `dsh-social-card` | `dsh-external-dsh-social-card-0.0.1.tgz` | social cards: image sets, 21:9 + 1:1 cover pairs, Live Photo plates |
| `dsh-academic-research` | `dsh-external-dsh-academic-research-0.0.1.tgz` | research pipeline: 10-stage state machine, five-seat review panel, integrity protocol |
| `dsh-pentagi` | `dsh-external-dsh-pentagi-0.0.1.tgz` | multi-agent penetration brain: 13 roles, seven-phase flow, adviser re-routing, knowledge base |
| `dsh-vuln-remediator` | `dsh-external-dsh-vuln-remediator-0.0.1.tgz` | vulnerability remediation: discovery, risk scoring, virtual patches, fix plans |
| `dsh-reverse-skill` | `dsh-external-dsh-reverse-skill-0.1.0.tgz` | reverse-engineering workflow: routing, confidence bands, evidence chain, validated gate |
| `dsh-office-docs` | `dsh-external-dsh-office-docs-0.0.1.tgz` | Office documents: build and extract PDF, DOCX, PPTX, XLSX; render pages; convert formats |
Install the body kernel plus the memory organ:
```powershell
$rel = "https://github.com/1420079678-ctrl/agent-body/releases/download/v0.1.1"
dsh plugin --profile web add "$rel/dsh-external-dsh-organism-0.1.1.tgz"
dsh plugin --profile web add "$rel/dsh-external-dsh-cortex-0.1.1.tgz"
```
The rest of the catalog needs a host-side toolchain (a compiler, a browser, or external binaries) and is installed with its
replay script instead — see the catalog section below.
### Or work from source
This repository is the **organ layer** — it does not vendor the host. Two pieces, both public:
```powershell
# 1) the host: DeepSeek Harness itself
git clone https://github.com/deepseek-ai/deepseek-harness.git
# follow that repository's build/run instructions; note where the checkout lives
# 2) the organs: this repository
git clone https://github.com/1420079678-ctrl/agent-body.git
```
Point the tooling at your host checkout, then bring an organ online. Each organ ships a **replay script** that links its
dependencies, compiles it, runs its offline regression, and tells you how to inject it:
```powershell
$env:DSH_CHECKOUT = "C:\path\to\deepseek-harness"
pwsh -File agent-body\workspace\plugins\dsh-organism\scripts\replay-organism.ps1
```
Replay scripts are idempotent — re-run them after a host upgrade to restore the organ. The same shape works for every
organ in the catalog (`replay-cortex.ps1`, `replay-web-crawl.ps1`, …).
Once injected, the first three commands to try:
```
body_status # full vitals: organs, fatigue, homeostasis, integrity self-check
body_map # the anatomy — which organ owns which capability
body_nerve action=send text="<your command>" # route a command before executing it
```
---
## Verify it yourself
Everything below runs offline and deterministically — **no network, no model, no API key, nothing installed**.
```bash
npm run check # the gate CI runs: constant tables + catalog + tests + benchmark, all in one
npm run demo # five-minute end-to-end: command → impulse → dispatch → execute → attribute → reflex
npm run bench # regenerate the token benchmark (writes benchmarks/results/REPORT.md)
```
`npm run check` is the honest one. It fails if the zero-dependency core drifts from the real kernel's constant tables,
if the organ catalog drifts from source, if any test fails, or if the benchmark moves off its committed baseline.
**All of it runs on a fresh clone with no `npm install`.**
### Organ regressions need the host runtime
```bash
npm run verify # repository health check (structure, JSON, links, secret hygiene)
npm run verify:organs # run every organ's offline regression
```
Each organ's suite loads its built `lib/`, which imports the host runtime (`@deepseek-ai/dsh-*`). On a bare clone that
does not resolve, so the runner reports **SKIP with the reason** — never a silent pass:
```
SKIP dsh-organism/smoke-test.mjs 缺少宿主运行时 @deepseek-ai/dsh-tools
SKIP dsh-web-crawl/selftest_local.py 缺少 Python 依赖 trafilatura
```
Inside an installed harness the same command runs them for real. Measured there:
| Organ | Command | Result (measured) |
| --- | --- | --- |
| `dsh-organism` | `node scripts/smoke-test.mjs` | **179 passed / 0 failed** (familyOf, evalCondition, decay, pruning, token estimation, gating contract) |
| `dsh-cortex` | `node scripts/smoke-test.mjs` | **58 passed / 0 failed** (tokenization, card mining, noise reduction) |
| `dsh-zero-residence` | `node scripts/smoke-test.mjs` | **16 passed / 0 failed** (pointer manifest, ledger math, exact-ID recall) |
| `dsh-war-bridge` | `node scripts/smoke-test.mjs` | **17 passed / 0 failed / 1 skipped** (the IDA chain skips without a sample PE) |
| `dsh-web-crawl` | `python scripts/selftest_local.py` | 46 offline assertions (extraction, magic‑byte routing, cascade) |
Two of these were red when this table was first written, and the failures were real:
- `dsh-zero-residence` reported **13 passed / 3 failed**. Root cause: session lookup matched by *substring* and returned
whichever directory the filesystem happened to yield first, so `computeLedger('session-a')` could read
`session-a-extra`'s log and report `T = 0` instead of `T = 2`. Fixed with exact-match-first resolution.
- `dsh-war-bridge` reported FAIL whenever no sample PE was present, because "skip" was recorded as a failed assertion —
and `process.exit()` racing an in-flight `AbortSignal.timeout` tripped a libuv assertion that turned a green run into
exit code 1. Both are fixed; a skip now says so.
The core packages are covered by `npm test`: **30 assertions, 0 failures**, including a **parity suite** that compares
the dependency-free port against the real kernel — exact agreement on `familyOf`/`tissueOf` across all 256 tool names,
`innervate` across 5 commands, `evalCondition` across 27 combinations, and `attributeFailure` across 8 error strings.
When no kernel build is present, parity **skips loudly** rather than passing quietly.
Runtime state is observable too — vitals, wounds, synapses and the pulse stream are all first‑class data:
```
body_status → organs · capabilities claimed · heartbeat #31 @15s · heal rate 100%
body_heal → 21 wounds healed · 0 chronic
body_pulse → last nerve impulses, reflex fires, homeostasis alerts
```
---
## Write your own organ
An organ is a declaration plus optional hooks. The SDK validates it where you write it and throws with a field path,
instead of failing silently at runtime:
```js
import { defineOrgan, injectedSource } from './packages/organ-sdk/src/index.mjs'
export default defineOrgan({
id: 'paper_reader',
label: 'Paper reading (literature)',
tier: 'professional',
group: 'memory',
purpose: 'turn one paper into searchable cards',
capabilities: ['paper_fetch', 'paper_digest'],
permissions: ['net:http'],
signals: ['tools/result'],
handles: ['network', 'timeout'],
fallback: ['hippocampus'],
})
```
Two rules that save a debugging session:
- **Inject with a producer-owned source kind.** `injectedSource('@you/paper-reader')` returns
`{ kind: 'plugin:@you/paper-reader' }` — the one shape both session-format generations accept. The retired
`{ kind: 'plugin', plugin: ... }` wrapper stops a whole turn on a format-V4 host; the measurements are in
[`docs/session-format-v4-compat.md`](docs/session-format-v4-compat.md).
- **Declare what you handle.** An organ that says it handles `timeout` gets that failure routed to it; one that stays
silent gets a compensating neighbour through `fallback` instead.
What your change has to pass:
```bash
npm run check # constants + catalog + message sources + tests + benchmark
npm run verify # structure, JSON, links, secret hygiene
```
The full contract is in [`docs/ORGAN_SDK.md`](docs/ORGAN_SDK.md), and wiring an organ into a real host is in
[`docs/host-adapter.md`](docs/host-adapter.md).
---
## When not to use this
Being straight about this is cheaper for both of us:
- **You do not run DeepSeek Harness.** This is a plugin layer for one specific host, not a standalone agent framework.
- **You want a frozen third-party API.** The host's session format changed once already and broke every plugin that
wrote to it; this repository tracks that host, so it moves when the host moves.
- **You need a supported product with an SLA.** v0.1.1 is a working release with reproducible gates, and the catalog is
curated against one development install — expect rough edges outside the paths that install exercises.
- **You need Linux/macOS parity today.** Several organs carry Windows-specific hardening (hidden-window spawning, ACL
sandbox compatibility). CI exercises the zero-dependency core on Linux; the organs less so.
---
## FAQ
**Do I need the host to try it?** No. `npm run demo` runs a real command → impulse → dispatch → execute → attribute →
reflex chain against the committed corpus with nothing installed and no API key.
**Does any of it call the network or a model?** Not in the part you can verify: `npm run check` installs nothing and
calls nothing. Model calls only happen where an organ asks the host to reason.
**What happens if I delete an organ?** Capability is lost, the body is not: the nervous system, heartbeat pump,
directive layer, reflex engine, anatomist and impulse dispatch depend on no single organ, and `body_call` compensates
with the online organ whose capability overlaps most. `body_organ action=integrity` prints that self-check.
**Why do the organ numbers differ between places?** There are three numbers and they measure different things:
**26 organ identities** in [`catalog/organs.json`](catalog/organs.json) (the anatomy — `prefrontal`, `hands`, `cortex`, …),
**24 plugin packages** under `workspace/plugins/` that realise them (23 in the catalog table above plus the legacy
`dsh-crawl4ai`, kept for rollback), and **64 catalog capabilities** declared in that file. `body_status` reports a
fourth, larger pair on a live install — 43 organs and 332 capabilities — because it counts every plugin the running
harness has mounted, including ones this catalog does not name. Every number is labelled wherever it appears, and
`npm run verify` prints the package count it derives from disk so it can be checked rather than trusted.
**How is the 84.7% token figure measured?** It is tool-schema tokens only — the sum of every tool definition's name,
description and parameters against what the first turn can see after gating — on 48 representative commands, in the
cold-start setting (intent decides the visible set, no local history). That is the lower bound and the only
independently reproducible setting. The live setting, with history, **declines as the body grows**: measured
**58%** on 2026‑09‑25 (127 of 332 visible), 74.25% on the 2026‑09‑11 snapshot. The whole series, and why it declines,
is in [Token economy](#token-economy); the raw data is in
[`benchmarks/results/REPORT.md`](benchmarks/results/REPORT.md).
---
## Contributing
The most useful contribution right now is **another organ** — the contract is small enough to read in one sitting.
Looking for somewhere to start? The
[**`good first issue`**](https://github.com/1420079678-ctrl/agent-body/labels/good%20first%20issue) label collects
scoped tasks with a stated acceptance test.
1. On a fresh clone, `npm run demo` and `npm run check`. If those are not green, that is a bug report worth filing on
its own.
2. Copy the closest organ under `workspace/plugins/`, declare yours with `defineOrgan`, and add its entry to
`catalog/organs.json` — the catalog is generated, and `npm run check:catalog` fails on drift rather than letting the
two diverge.
3. Ship an **offline regression** with it. Only **5 of the 24 packages have one today** — `npm run verify` prints the
list of the 18 that do not — and `npm run verify:organs` reports SKIP with a reason rather than passing quietly when
the host runtime is absent. Writing those tests, or an organ that brings its own, is the most useful contribution
right now.
4. Open a PR describing what the organ does, which system it belongs to, and the exact commands you ran.
Questions and discussions are welcome in the harness's
[**Show Your Plugins!**](https://github.com/deepseek-ai/deepseek-harness/discussions/7555) thread or as an issue here.
---
## Repository layout
```
agent-body/
├─ packages/
│ ├─ organ-core/ zero-dependency organ core: constant tables, gating, contract, host adapter
│ └─ organ-sdk/ defineOrgan() / defineReflex() — the authoring surface
├─ benchmarks/ reproducible token benchmark + frozen corpus + CI baseline
├─ catalog/ organs.json — tiers, permissions, failure handling (generated, drift-checked)
├─ examples/ the five-minute quickstart demo
├─ docs/ organ SDK guide · host adapter guide
├─ workspace/
│ └─ plugins/ the organs — one plugin per organ, each with src/ + lib/ (offline regression where one exists)
├─ scripts/ verify-repo.mjs · run-organ-regressions.mjs · run-tests.mjs
├─ .github/ CI (repo-check on Windows + Linux) and issue templates
├─ ARCHITECTURE.md full system architecture
├─ ROADMAP.md what the next six months are for
└─ README.zh-CN.md 中文文档
```
Runtime state (vitals, synapses, memory cards, pulse stream) lives in your harness data home, never in this repository.
---
## Roadmap
- [ ] Publish the organ contract as a standalone SDK so third‑party plugins can declare organs in a few lines
- [ ] One‑command organ installer that links, builds and registers every organ against a given host checkout
- [ ] Cross‑body sync: export learned synapses / reflexes / skills and import them into another install
- [ ] Sleep‑time model training: let consolidation propose new reflexes for review instead of authoring them directly
- [ ] Web panel for the anatomy: live organ map, wound ledger, pulse stream
---
## Notes
- **Licensing.** MIT for this project's own work ([LICENSE](LICENSE)). Plugins ported from upstream projects and the
host itself keep their own terms — see [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).
- **Local‑first.** The harness GUI binds `127.0.0.1` — it can read files and execute commands on the machine it runs on.
Do not expose it to a network or put it behind a public reverse proxy.
- **Windows‑first.** Developed and measured on Windows 11 · Node 24 · PowerShell 7. Several organs ship
Windows‑specific hardening (hidden‑window spawning, ACL sandbox compatibility). Linux/macOS paths exist but are less
exercised.
- **Numbers here are measured**, not aspirational. Figures marked *measured* were read from a live install; runtime
counts (organs, capabilities) are yours to reproduce with `body_status`.
<div align="center">
---
****Six kernels. 26 cataloged organ identities across 8 systems, realised by 24 plugins. One heartbeat.**
If that's the kind of plugin platform you want, the architecture is all in [ARCHITECTURE.md](ARCHITECTURE.md).
⭐ **If this body has a pulse for you, [star this repo](https://github.com/1420079678-ctrl/agent-body/stargazers)** — it costs
nothing and is the single easiest way to help others find the project.
**Community.** Discussed in the [LINUX DO](https://linux.do/) community — a friendly Chinese-speaking
developer forum where this project was shared and where questions get answered.
<img src="https://api.star-history.com/svg?repos=1420079678-ctrl/agent-body&type=Date" alt="Star history chart" width="100%">
</div>
连接信息
猜你喜欢
暂无推荐服务器