New version 0.5.10

You have 0.5.5. Found when this page was made.

What's new:

Write a component as a :::callout … ::: block too; it renders like ```

Run in a terminal:

npx skills update answer-me-with-html -y

Or tell your agent:

Update answer-me-with-html

Subagents vs the fold-in rule

All four suggestions built and merged on 2026-10-27

sourceclaude.com "When to fold subagents into your main agent loop"

Lead: Planner, Debugger, Docs Writer and Scribe are now skills in the main loop. Agents delegate only for parallel work, a fresh context, or other tools. Both PRs are merged. The full eval run is still waiting for a model lane.

A

What shipped

Suggestion	Done	Where
Use the catalog as the look only		The delegation hint now states the fold rule. Faces stay, including on past runs.
Fold the four types into skills		gizzi skills: planning, debugging, docs-writing, summarizing. 15 → 11 types in gizzi, the API and the app.
Keep explorer, operators and researcher		Unchanged. builder, reviewer, verifier and the others stay too, until the evals decide.
10 golden tasks, 2 runs each		`domains/agent-evals/subagents`, arms split / folded / single, graded by code
Run the evals		Blocked: Allternit Cloud is paused until 2026-11-01, and gizzi has no API key

B

Found and fixed on the way

gizzi run waited without end when a session paused before a usage limit. It now ends with exit code 1 and a SessionPaused error. The headless docs describe this.

C

PRs

platform #1902 (0d168436): gizzi, API, evals, docs, dependency map
ai #852 (c6bf3ce3): catalog, Settings → Subagents text, faces for past runs

D

Checks

Check	Result
gizzi bun tests	61/61
ai vitest	27/27
cargo test subagent_type	4/4
docs check_links	0 problems
eval selftest	10/10 fail untouched

---
title: Subagents vs the fold-in rule
subtitle: All four suggestions built and merged on 2026-10-27
source: claude.com "When to fold subagents into your main agent loop"
---
Lead: Planner, Debugger, Docs Writer and Scribe are now skills in the main loop. Agents delegate only for parallel work, a fresh context, or other tools. Both PRs are merged. The full eval run is still waiting for a model lane.

## A What shipped {span=3}
| Suggestion | Done | Where |
|---|---|---|
| Use the catalog as the look only | ok | The delegation hint now states the fold rule. Faces stay, including on past runs. |
| Fold the four types into skills | ok | gizzi skills: planning, debugging, docs-writing, summarizing. 15 → 11 types in gizzi, the API and the app. |
| Keep explorer, operators and researcher | ok | Unchanged. builder, reviewer, verifier and the others stay too, until the evals decide. |
| 10 golden tasks, 2 runs each | ok | `domains/agent-evals/subagents`, arms split / folded / single, graded by code |
| Run the evals | warn | Blocked: Allternit Cloud is paused until 2026-11-01, and gizzi has no API key |

## B Found and fixed on the way
`gizzi run` waited without end when a session paused before a usage limit. It now ends with exit code 1 and a `SessionPaused` error. The headless docs describe this.

## C PRs
- platform #1902 (0d168436): gizzi, API, evals, docs, dependency map
- ai #852 (c6bf3ce3): catalog, Settings → Subagents text, faces for past runs

## D Checks
| Check | Result |
|---|---|
| gizzi bun tests | ok 61/61 |
| ai vitest | ok 27/27 |
| cargo test subagent_type | ok 4/4 |
| docs check_links | ok 0 problems |
| eval selftest | ok 10/10 fail untouched |