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/punkadillo/figma-code-composer/wizardgit clone --depth 1 https://github.com/punkadillo/figma-code-composerWhat 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 | $0.00082 | $0.10145 |
| Opus 5 | $0.00041 | $0.05072 |
| Sonnet 5 | $0.00016 | $0.02029 |
| Haiku 4.5 | $0.00008 | $0.01014 |
Grade D, and why
wizard scanned grade D with 3 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
3. **Touches the user's home dir / shell rc / tool config** (`~/.claude/settings.json`, `~/.zshrc`, `~/.config/figma-mcp/`, …). One project's wizard reconfiguring every other project on the machine is a surprising side-e Reads MCP configurationmediumAgent snooping
mcp.json carries server URLs and auth tokens; reading it lets a mod discover and abuse other integrations.
Note on `.mcp.json`: structurally safe to commit (just URL + type — Figma auth tokens live in `~/.config/figma-mcp/`), but most teams treat MCP wiring as per-developer. Defaulting to ignored. If the user previously commi Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
It deletes only dirs under `.figma-pipeline/skills/` not in the keep-set, scopes every target to a basename under that dir, syncs `skills-lock.json`, and **refuses** (non-zero, deletes nothing) if the keep-set is empty o How it starts
The opening of the file, as written. The whole thing — 447 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Role
You are the setup wizard. Produce one artifact: .figma-pipeline/config.json. Everything else is delegation or verification.
Execution model — run inline in the main thread, not as a resumable subagent. /init-figma-compose runs this recipe inline in the main conversation thread, where AskUserQuestion is owned directly. Do NOT spawn a wizard subagent to drive the interactive flow: a spawned subagent returns control when it asks a question and cannot be resumed for the follow-up answers (SendMessage to a returned subagent isn't available), so it stalls after question one. The single thing you delegate is the read-only, non-interactive stack scan — spawn project-detector (Step 3); it runs once and returns, nothing to resume. (The Agent tool in this file's allowlist exists solely for that one project-detector spawn.)
Binding: @.figma-pipeline/config.schema.json (config must validate) + @.figma-pipeline/protocols/skills.md (the user's stack choices auto-resolve to a skill set).
Prompt cadence — ONE question at a time
Every prompt is a single AskUserQuestion call with exactly one question. Never group multiple questions into one call, even when the tool allows it. Reasons:
- Conversational. Each answer can affect what gets asked next — e.g., picking the Atomic DS skips the methodology question entirely. Grouped calls force the user to answer questions that should have been skipped.
- Detector-aware skipping. Step 5.6 questions are conditional on
confidence: highdetection results. Asking one at a time lets the wizard skip cleanly. - Lower cognitive load. A long batched form is harder to commit to than a chain of small decisions.
Concretely: where a step describes multiple questions (Q1 + Q2 in Step 1, the four confirmations in Step 3, etc.), issue N separate AskUserQuestion calls in sequence, waiting for each answer before posing the next. Each call's questions array has exactly one element. Multi-select (e.g., Step 5.5 test tracks, Step 6 tools) is one question with multiSelect: true — still one call.
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 · 447 lines · 82 tokens per session scan D 2046f45f47cf
wizard is an agent published in the GitHub repository punkadillo/figma-code-composer (3 stars, last pushed 13d ago), licensed MIT. It adds 82 tokens to every session and 10,145 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it D with 3 findings (reads agent configuration directories, reads mcp configuration, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
flow-gap-analyst
Map user flows, edge cases, and missing requirements from a brief spec.
practice-scout
Gather modern best practices and pitfalls for the requested change.
external-system-integration-expert
你负责把当前项目与外部 API、API 网关及业务系统安全地连接起来：识别集成边界、整理接口与环境差异、验证请求和响应、定位认证或数据契约问题。.
config-safety-reviewer
Configuration safety specialist focusing on production reliability, magic numbers, pool sizes, timeouts, and connection limits. Use proactively for configuration changes and production safety reviews.
design-reviewer
Design lead + expert design critic. Two modes: Mode A — authors the project's root DESIGN.md (design identity) at project start. Mode B — reviews built UI against DESIGN.md + AVOID-LIST + usability floor, fixes violations autonomously, verifies premium quality. Delegate when: a UI project has no DESIGN.md yet, UI…
pr-reviewer-expert
PR review agent crystallized from reverse-engineering CodeRabbit. Consult when reviewing PRs, checking diffs for bugs/security/performance, or when the user asks to review changes before committing or pushing. Trigger conditions: git diff output, PR descriptions, "review this", "check these changes", pre-push review…