agent-builder

Name: agent-builder
Author: testdouble

122stars12forks

How to install agent-builder?

testdouble/han/agent-builder

$npx -y skills add testdouble/han --skill agent-builder

Installs into the current project.

Summary

Builds a new Claude Code agent (subagent) from scratch through a relentless, evidence-based interview that walks the agent's design tree decision-by-decision — entity fit, domain focus and vocabulary, role identity, anti-patterns, description, model tier, tools, and self-containm

Files · 1

View on GitHub

SKILL.md

1	## Guidance Location
2
3	The authoritative agent-authoring guidance ships in this plugin. Read the
4	specific document a decision needs, when that decision is on the table — never
5	read them all up front, because that defeats progressive disclosure and burns
6	context on guidance the current agent does not touch.
7
8	- Plugin-building guidance root: `${CLAUDE_PLUGIN_ROOT}/skills/guidance/references/`
9	- Agent-specific guidance: `${CLAUDE_PLUGIN_ROOT}/skills/guidance/references/agent-building-guidelines/`
10
11	Map from decision to governing document (read just-in-time):
12
13	\| Decision on the table \| Read \|
14	\|---\|---\|
15	\| Agent vs. skill vs. hook; one role (generate or evaluate) \| `plugin-entity-taxonomy.md`, `agent-building-guidelines/agent-domain-focus.md` \|
16	\| Domain focus, vocabulary, role identity, anti-patterns \| `agent-building-guidelines/agent-domain-focus.md` \|
17	\| The `description` field (four components, boundaries, length) \| `agent-building-guidelines/agent-description-length.md`, `skill-building-guidance/skill-description-frontmatter.md` \|
18	\| Model tier (`opus` / `sonnet` / `haiku` / `inherit`) \| `agent-building-guidelines/agent-model-selection.md`, `specialization-and-model-selection.md` \|
19	\| Self-containment — no references, scripts, or context injection \| `agent-building-guidelines/agent-external-files.md` \|
20	\| Which frontmatter fields are valid (and which plugins ignore) \| `agent-building-guidelines/agent-external-files.md` \|
21	\| Degraded environments (no git, missing tools) \| `agent-building-guidelines/graceful-degradation.md` \|
22	\| Whether this agent is justified at all; how it gets dispatched \| `agent-building-guidelines/multi-agent-economics.md`, `skill-building-guidance/agent-dispatch-namespacing.md` \|
23	\| New plugin needed (plugin.json, marketplace.json) \| `claude-marketplace-and-plugin-configuration/` and `templates/` \|
24
25	## Operating Principles
26
27	- Interview relentlessly, but explore first. Interview the user relentlessly
28	about every aspect of the agent until you reach a shared understanding. Walk
29	down each branch of the design tree, resolving dependencies between decisions
30	one-by-one. **If a question can be answered by exploring the repository — the
31	target plugin's existing agents, sibling descriptions, the skills that would
32	dispatch this agent, conventions, the guidance documents above — explore
33	instead of asking.** Only surface questions that genuinely require the user's
34	judgment.
35	- Ask one question at a time. Never batch questions. Settle one decision,
36	let its answer resolve dependent decisions, then ask the next. Later answers
37	routinely make earlier questions moot.
38	- Recommend, then ask. For every question surfaced to the user, provide a
39	recommended answer with rationale grounded in evidence (existing agents,
40	conventions, the guidance, the user's stated goal). The user can accept,
41	amend, or redirect.
42	- Apply guidance as you go, then verify at the end. Consult the governing
43	document when a decision is on the table (Step 4), and run a full
44	guidance-conformance pass over the finished agent at the end (Step 6). The
45	interview gets each decision approximately right; the review pass makes the
46	artifact correct.
47	- An agent is self-contained. Unlike a skill, an agent is a single flat
48	`.md` file. No `references/` folder, no `scripts/` folder, no `` !`command` ``
49	context injection. Everything the agent needs is inlined in its body.
50
51	# Build an Agent
52
53	## Step 1: Capture the Request and Confirm It Is an Agent
54
55	Read the user's argument and the conversation to extract what the agent should
56	do. If the request is too thin to start (for example, just "build an agent"),
57	ask the user for one or two sentences on the agent's domain and what it produces
58	— nothing else yet.
59
60	Confirm the entity type before anything else. Read
61	`${CLAUDE_PLUGIN_ROOT}/skills/guidance/references/plugin-entity-taxonomy.md` and
62	apply its decision heuristic. An agent is the thinking layer: it applies
63	contextual judgment, taste, and discernment ("Does this require reasoning about
64	context?" → agent). If the work is a deterministic, flowchartable process, it is
65	a skill — stop and recommend `skill-builder`. If it fires automatically on an
66	event, it is a hook.
67
68	**Confirm t

People who liked this also viewed

6 picks

Type

agent-builder

122stars12forks

View on GitHub

How to install agent-builder?

testdouble/han/agent-builder

$npx -y skills add testdouble/han --skill agent-builder

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	## Guidance Location
2
3	The authoritative agent-authoring guidance ships in this plugin. Read the
4	specific document a decision needs, when that decision is on the table — never
5	read them all up front, because that defeats progressive disclosure and burns
6	context on guidance the current agent does not touch.
7
8	- Plugin-building guidance root: `${CLAUDE_PLUGIN_ROOT}/skills/guidance/references/`
9	- Agent-specific guidance: `${CLAUDE_PLUGIN_ROOT}/skills/guidance/references/agent-building-guidelines/`
10
11	Map from decision to governing document (read just-in-time):
12
13	\| Decision on the table \| Read \|
14	\|---\|---\|
15	\| Agent vs. skill vs. hook; one role (generate or evaluate) \| `plugin-entity-taxonomy.md`, `agent-building-guidelines/agent-domain-focus.md` \|
16	\| Domain focus, vocabulary, role identity, anti-patterns \| `agent-building-guidelines/agent-domain-focus.md` \|
17	\| The `description` field (four components, boundaries, length) \| `agent-building-guidelines/agent-description-length.md`, `skill-building-guidance/skill-description-frontmatter.md` \|
18	\| Model tier (`opus` / `sonnet` / `haiku` / `inherit`) \| `agent-building-guidelines/agent-model-selection.md`, `specialization-and-model-selection.md` \|
19	\| Self-containment — no references, scripts, or context injection \| `agent-building-guidelines/agent-external-files.md` \|
20	\| Which frontmatter fields are valid (and which plugins ignore) \| `agent-building-guidelines/agent-external-files.md` \|
21	\| Degraded environments (no git, missing tools) \| `agent-building-guidelines/graceful-degradation.md` \|
22	\| Whether this agent is justified at all; how it gets dispatched \| `agent-building-guidelines/multi-agent-economics.md`, `skill-building-guidance/agent-dispatch-namespacing.md` \|
23	\| New plugin needed (plugin.json, marketplace.json) \| `claude-marketplace-and-plugin-configuration/` and `templates/` \|
24
25	## Operating Principles
26
27	- Interview relentlessly, but explore first. Interview the user relentlessly
28	about every aspect of the agent until you reach a shared understanding. Walk
29	down each branch of the design tree, resolving dependencies between decisions
30	one-by-one. **If a question can be answered by exploring the repository — the
31	target plugin's existing agents, sibling descriptions, the skills that would
32	dispatch this agent, conventions, the guidance documents above — explore
33	instead of asking.** Only surface questions that genuinely require the user's
34	judgment.
35	- Ask one question at a time. Never batch questions. Settle one decision,
36	let its answer resolve dependent decisions, then ask the next. Later answers
37	routinely make earlier questions moot.
38	- Recommend, then ask. For every question surfaced to the user, provide a
39	recommended answer with rationale grounded in evidence (existing agents,
40	conventions, the guidance, the user's stated goal). The user can accept,
41	amend, or redirect.
42	- Apply guidance as you go, then verify at the end. Consult the governing
43	document when a decision is on the table (Step 4), and run a full
44	guidance-conformance pass over the finished agent at the end (Step 6). The
45	interview gets each decision approximately right; the review pass makes the
46	artifact correct.
47	- An agent is self-contained. Unlike a skill, an agent is a single flat
48	`.md` file. No `references/` folder, no `scripts/` folder, no `` !`command` ``
49	context injection. Everything the agent needs is inlined in its body.
50
51	# Build an Agent
52
53	## Step 1: Capture the Request and Confirm It Is an Agent
54
55	Read the user's argument and the conversation to extract what the agent should
56	do. If the request is too thin to start (for example, just "build an agent"),
57	ask the user for one or two sentences on the agent's domain and what it produces
58	— nothing else yet.
59
60	Confirm the entity type before anything else. Read
61	`${CLAUDE_PLUGIN_ROOT}/skills/guidance/references/plugin-entity-taxonomy.md` and
62	apply its decision heuristic. An agent is the thinking layer: it applies
63	contextual judgment, taste, and discernment ("Does this require reasoning about
64	context?" → agent). If the work is a deterministic, flowchartable process, it is
65	a skill — stop and recommend `skill-builder`. If it fires automatically on an
66	event, it is a hook.
67
68	**Confirm t

People who liked this also viewed

6 picks

Type