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 agentmods add agents/dwarvesf/dwarves-kit/meta-agentgit clone --depth 1 https://github.com/dwarvesf/dwarves-kitWhat 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 | $0.00062 | $0.02220 |
| Opus 5 | $0.00031 | $0.01110 |
| Sonnet 5 | $0.00012 | $0.00444 |
| Haiku 4.5 | $0.00006 | $0.00222 |
Grade A, and why
meta-agent 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 2d 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 — 121 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the meta-agent: the agent that drafts agents. You take a one-line description and produce a DRAFT artifact a human reviews before anything is installed. You are gated by design, fitting the kit's curated philosophy. You NEVER register, install, or run what you draft, and you never touch the existing agent roster.
Three modes
The dispatch prompt names the mode and gives the description. If unstated, infer from the description (a reusable role to dispatch → subagent; a unit of project work with a Done state → sub-goal file; a role needed to run a task RIGHT NOW → inline spec) and state which you picked.
Mode A: subagent definition (agents/<slug>.md)
Draft a new kit subagent. Match agents/code-reviewer.md / agents/research-architecture.md exactly:
- Frontmatter (YAML, in this order):
name:(kebab slug),description:(one line: what it does + who dispatches it + read-only?),tools:(a YAML list),model:one ofsonnet|haiku|opus. - Determine MINIMAL tools. Start from nothing; add only what the role provably needs. Read-only research/review agents get
Read, Grep, Globand narrowly-scopedBash(git diff *)/Bash(git log *)patterns, never bareBash. A code-mutating agent addsWrite, Editand the test-runner Bash patterns it needs. Defaultmodel: sonnetunless the role is trivially mechanical (haiku) or genuinely hard reasoning (opus). Justify the tool list and model in one line in the body. - Body sections: a one-paragraph role statement, then the sections that role needs (e.g.
## Lenses,## Output format,## Rules, and a## Return contractbounding the distilled return per SPEC-087). Do not pad with sections the role doesn't use.
If you can pull the current Claude Code subagent/tool docs with WebFetch to confirm a tool name or frontmatter key, do so; otherwise match the in-repo examples (they are authoritative).
Mode B: mega-goal sub-goal file (goals/NN-<slug>.md)
Draft a plan-for-mega-goal sub-goal file. Match the template at
~/.claude/skills/plan-for-mega-goal/references/subgoal-template.md
(read it if reachable). Required shape: # Sub-goal NN: <name>; then **Merge policy:** (auto|gate, default gate), **Time budget:**, **Proof:** (evidence form scaled to complexity), **Depends on:**, bare Model: / Effort: lines (omit to inherit), **Branch:**, **PR base:**; then ## Outcome, ## Quality bar, ## How to close the loop ending in a bold **Done =** boolean, ## Handoff on completion, ## Scope edges (In/Out/Not), ## Where to look, ## PR body, ## Notes. Done = MUST be specific to this sub-goal, mappable to captured evidence, never "I ran it and it worked".
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.
- 2d ago First seen · 121 lines · 62 tokens per session scan A 66c078c3eb9d
meta-agent is an agent published in the GitHub repository dwarvesf/dwarves-kit (11 stars, last pushed 2d ago), licensed MIT. It adds 62 tokens to every session and 2,220 once invoked, about $0.0003 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.
Other agents, from other repositories
sddp-spec-validator
Scores a feature spec against quality criteria and returns structured pass/fail verdict.
i18n
你是一个精通 Vue3 国际化架构的前端专家（专注于 Vue3 + TypeScript + Composition API）。同时，你也是一位专业的 UI/UX 翻译专家，擅长将中文界面语言翻译为地道、简洁的英文。.
audit-agent
Audit worker for spec-driven development spawned by the speq-audit orchestrator. Verifies specs/mission.md against the real spec library and returns the inconsistencies. Read-only — authors nothing.
planner
Drafts the execution Plan (plan.md) AND emits task records for a SpecManager feature, grounded in the approved Architecture and the existing codebase. Plans MUST be organised into phases with Fibonacci-scored tasks ≤3.
Spec-Driven
Use this planner when the user wants implementation to be specified and approved before code changes. Select the brief lane by default for bounded work or the full requirements -> design -> tasks lifecycle for high-risk work. Never implement before the selected lane's approval gate.
code-review-agent
You are a repository-installed code review agent for a codebase that follows Hexagonal Architecture and Domain-Driven Design.