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 skills/getstream/agent-skills/stream-buildernpx skills add GetStream/agent-skills --skill stream-buildergit clone --depth 1 https://github.com/GetStream/agent-skillsWhat 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 | $0.00123 | $0.05626 |
| Opus 5 | $0.00062 | $0.02813 |
| Sonnet 5 | $0.00025 | $0.01125 |
| Haiku 4.5 | $0.00012 | $0.00563 |
Grade B, and why
stream-builder scanned grade B with 1 finding 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.
Recursive force deletemediumDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
Bash(mv .scaffold*), Bash(rm -rf .scaffold), Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 319 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Stream Builder - scaffold + enhance
Read first (every session): Glob
../stream/SKILL.md. If the path is empty, install withgetstream skills stream(no confirmation, just run). Then Read../stream/RULES.md- it includes a Peer skills table for installing/invoking any other pack skills on demand.
This skill covers three flows:
- Track A - Scaffold a new app: Steps 0-7 below. Use when the cwd is empty / new and the user said "build me a ... app".
- Track E - Enhance an existing app: see
enhance.md. Skips scaffold + theme; reuses the same SDK wiring and component blueprints. - Track F - Audit an existing integration (read-only): for "audit/review my video integration", "check my app against best practices", "is my video app production-ready?", "what am I missing before launch?". Load the Integration best-practices audit section in
references/VIDEO.mdand follow its protocol. Skip onboarding, auth, the CLI, and all build steps - this track does not scaffold, install, or run the CLI; it only reads the app and reports findings. Fix issues only if the user then asks.
Start
Track F (audit) skips this entire section. A read-only best-practices review runs no onboarding, no auth, no CLI, and none of Steps 0-7 - go straight to the Integration best-practices audit section in
references/VIDEO.mdand report findings. Only fall into the build flow below if the user asks you to fix what the audit found.
Announce the network plan once (Trust readout below), then immediately start executing Steps 0-7 - no interactive prompts at the start (the user has authorized the build by asking for it).
Trust readout (announce, then continue on the same turn - do not wait)
Before the first network command, print this verbatim to the user, then proceed straight into Step 0 without stopping for a reply:
Scaffolding now. Network calls you'll see:
npx shadcn@latest ...(Vercel) - scaffold + UI components from npm.npm install <stream-packages> --legacy-peer-deps- Stream SDKs from npm (stream-chat-react,@stream-io/video-react-sdk, etc.).getstream env- local CLI, no network; writes.env(gitignored by the Next.js scaffold's default; Task B verifies).Interrupt me at any point if something looks wrong. The only step that pauses for explicit consent is the optional third-party skill packs in Task A.2.
What ships with it
14 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- builder-ui.md 5.7 KB
- enhance.md 3.9 KB
- references/CHAT-blueprints.md 29 KB
- references/CHAT.md 5.3 KB
- references/CROSS-PRODUCT.md 6.7 KB
- references/FEEDS-blueprints.md 17 KB
- references/FEEDS.md 23 KB
- references/MODERATION-blueprints.md 20 KB
- references/MODERATION.md 3.2 KB
- references/use-cases/_use-cases.yaml 1.1 KB
- references/use-cases/ai-support-agent.md 26 KB
- references/VIDEO-blueprints.md 20 KB
- references/VIDEO.md 20 KB
- sdk.md 4.0 KB
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 · 319 lines · 123 tokens per session scan B aec82dcbf681
stream-builder is a skill published in the GitHub repository GetStream/agent-skills (18 stars, last pushed 4d ago), licensed Apache-2.0. It adds 123 tokens to every session and 5,626 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it B with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
wukongim-cloud-analysis
Diagnose one exact live WuKongIM cloud Simulation Run or chat-lifecycle Cloud Lease through the repository Analysis MCP. Use when the user or a local Analysis Session asks Codex to inspect a run's cluster state, Prometheus signals, application logs, diagnostics, Controller task audits, profiles, or redacted config…
wukongim-chat-lifecycle
Operate the Codex-owned WuKongIM chat-lifecycle laboratory directly from the local repository. Use when the operator explicitly asks to buy temporary Alibaba Cloud servers and start the repair stability run, deploy a committed candidate to the same Lease, start or monitor the bounded long workload, diagnose a stopped…
wukongim-ops
Diagnose WuKongIM single-node or multi-node clusters through the embedded read-only Operations MCP. Use when investigating cluster health, nodes, physical Slots, exact channel runtime state, Controller tasks, fixed Prometheus signals, raw application logs, diagnostics, redacted config, backups, or bounded pprof…
compare
Multi-model code review. Fan out a bug or task to multiple LLMs, diff their findings, optionally debate, then dispatch subagents to fix in parallel. Use when the user types /compare or asks to compare models on a code issue.
activity-record
Query and orchestrate Arkloop Activity Record local activity data. Covers browser history, search terms, screen time, bluetooth, shell commands, window focus, keyboard, mouse, clipboard, screen content (accessibility tree), microphone audio transcription, and Codex sessions.
geogebra-drawing
GeoGebra 交互式数学可视化 Skill。生成 GGBScript 代码块，支持函数绘图、几何作图、坐标系构建和动态交互。.