brainstorm — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited brainstorm (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Guided ideation pipeline that turns rough ideas into a validated requirements document. Alternates between conversation with the user and delegated research to keep main context lean.
/brainstorm [topic or idea]Examples:
/brainstorm "real-time notifications system"/brainstorm "mobile-responsive dashboard"/brainstorm "plugin marketplace monetization"NO agents. NO file reads. Just conversation.
Main's job:
Output: A numbered list of candidate ideas, each 1-2 sentences.
Context budget: This phase should use <500 tokens of main's context.
For EACH candidate idea from Phase 1, spawn research agents in parallel. Main does NOT read their raw output -- agents return a structured verdict.
Agents per idea (launch all in parallel):
| Agent | Question | Method | Verdict format |
|---|---|---|---|
feasibility | "Can our stack do this?" | Read architecture docs, relevant source files | FEASIBLE / RISKY / INFEASIBLE + 2-line reason |
prior-art | "How do serious players solve this?" | Web search for established approaches | 3 bullet points: who does it, how, what we can learn |
conflict-check | "Does this break or duplicate anything we have?" | Read architecture docs, grep for overlapping features | CLEAR / CONFLICT + what it conflicts with |
Agent prompt template:
You are a research agent. Answer ONE question and return a structured verdict.
QUESTION: [question]
IDEA: [idea description]
PROJECT: [brief context from Phase 1]
Research approach:
- [specific files to read or searches to run]
Return your verdict in EXACTLY this format:
VERDICT: [FEASIBLE|RISKY|INFEASIBLE] or [CLEAR|CONFLICT] or [bullet points]
REASON: [2-3 lines max, be specific, cite what you found]
Do NOT return raw file contents or search results. Only the verdict.Main receives: ~10 lines per idea (3 verdicts x ~3 lines each). For 4 ideas, that's ~40 lines. Context stays lean.
Present research findings as a summary table:
## Research Results
| Idea | Feasible? | Prior Art | Conflicts? | Recommendation |
|------|-----------|-----------|------------|----------------|
| Real-time updates | RISKY -- WebSocket complexity | Competitors use SSE | CLEAR | Worth it if we use SSE instead |
| Mobile dashboard | FEASIBLE -- CSS-only responsive | Every competitor has this | CLEAR | Low effort, high impact |Then ask the user:
For each surviving idea, resolve these 6 questions through conversation. Do NOT proceed to output until all 6 are answered for every idea.
Track internally -- check off as each is answered:
If the user hasn't addressed one, ask directly: "We haven't nailed down [X] yet -- any constraints there, or should I leave it open for the planner?"
Write the requirements doc to docs/brainstorms/{TOPIC}_REQUIREMENTS.md.
Format:
# [Topic] -- Requirements
> Brainstormed: [date]
> Status: Ready for /plan-features
## Constraints
- [Hard constraints from Phase 4]
## Features (priority order)
### 1. [Feature Name]
**Description:** [What, not how]
**Success criteria:** [Observable outcome]
**Out of scope:** [What this does NOT include]
**Research notes:** [Key findings from Phase 2 -- feasibility, prior art]
**Open decisions:** [Anything the planner needs to resolve]
### 2. [Feature Name]
[same structure]
## Deferred Ideas
- [Idea] -- deferred because [reason from Phase 3]
## Killed Ideas
- [Idea] -- killed because [reason from Phase 2/3]Tell the user: "Requirements doc written. Run /plan-features --doc docs/brainstorms/{TOPIC}_REQUIREMENTS.md to generate the execution plan."
Main's total context growth for a 4-idea brainstorm should be under 200 lines. This leaves room to run /plan-features and /exec-plan in the same session.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.