features/agents

features/agents is an agent for coding agents from plasma-ai/fractal. It costs 3 tokens per session (491 once invoked), scanned A, original, Apache-2.0.

Documentation for a system that runs coding agents as subprocesses, meaning separate programs controlled by another program, with support for hierarchical agent loops.

In plain words
What is it for?
Understanding supported agent backends, adding or selecting providers, running nested agent workflows, and handling sessions, models, effort settings, and streamed output.
Why use it?
It gives different agent providers one shared execution interface while keeping provider-specific code separate.

Agent

Part of the fractal plugin — 4 skills, 4 agents shipped together

Install

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.

agentmods
npx agentmods add agents/plasma-ai/fractal/_index
Clone the repo
git clone --depth 1 https://github.com/plasma-ai/fractal

Or install fractal, the plugin that ships this one along with the rest of its 4 skills, 4 agents.

Wrote 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.

agentmods badge for features/agents

README.md
[![agentmods](https://agentmods.dev/badge/agents/plasma-ai/fractal/_index.svg)](https://agentmods.dev/agents/plasma-ai/fractal/_index)
Your own site
<a href="https://agentmods.dev/agents/plasma-ai/fractal/_index"><img src="https://agentmods.dev/badge/agents/plasma-ai/fractal/_index.svg" alt="Measured on agentmods" height="20"></a>
Per session 3 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 491 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00003 $0.00491
Opus 5 $0.00002 $0.00246
Sonnet 5 $0.00001 $0.00098
Haiku 4.5 $0.00000 $0.00049

Measured 4d ago against content hash d03d8261d438, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

features/agents 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.

wiki/features/agents/_index.md · 50 lines

What it actually says

features/agents

[[features/_index|..]]

[[features/agents/extending|extending]]: How a new provider backend slots in: one new implementation module whose agent subclass declares its capabilities and overrides the private hooks, registered in the core agent module or injected from outside the package.

[[features/agents/models_and_effort|models_and_effort]]: Where a spawn's model and reasoning-effort overrides come from, how each provider backend spells them on its command line, how a backend resolves the model its own configuration defaults to, and how the served model is recorded off the stream.

[[features/agents/providers|providers]]: The supported agent backends, the registry in the core agent module that resolves a base command to its backend class, and the provider routes a backend may expose beside its vendor-native endpoint.


The agent execution surface. Fractal drives coding agents as subprocesses behind one base-class contract in fractal/core/agent.py, with one backend module per provider in fractal/impl/.

  • [[features/agents/providers]] — the five supported backends, the registry that resolves a base command to its backend class, capability declarations, and the provider routes (openrouter) beside the vendor-native endpoints.
  • [[features/agents/models_and_effort]] — where model and reasoning-effort overrides come from (node config keys and per-invocation overrides), each backend's CLI spelling of them, the configured-model fallback, and model preflight.
  • [[features/agents/extending]] — the core/impl seam: the subclass shape (capability attributes, private hooks, observability events) and the three registration points (registry entry, in-process register, deployment hook file).

Transcript access for a node's sessions is fronted by the project-files surface ([[features/files/_index|files]]); the run loop that consumes these agents lives in fractal/core/loop.py.

Changes

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.

  1. 4d ago First seen · 50 lines · 3 tokens per session scan A d03d8261d438

Subscribe to this mod's changes

features/agents is an agent published in the GitHub repository plasma-ai/fractal (704 stars, last pushed 4d ago), licensed Apache-2.0. It adds 3 tokens to every session and 491 once invoked, about $0.0000 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 2026-08-30.