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/yaodub/cast/manualnpx skills add yaodub/cast --skill manualgit clone --depth 1 https://github.com/yaodub/castWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/yaodub/cast/manual)<a href="https://agentmods.dev/skills/yaodub/cast/manual"><img src="https://agentmods.dev/badge/skills/yaodub/cast/manual.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.00022 | $0.00696 |
| Opus 5 | $0.00011 | $0.00348 |
| Sonnet 5 | $0.00004 | $0.00139 |
| Haiku 4.5 | $0.00002 | $0.00070 |
Grade B, and why
manual 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 4d 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.
Subtle steeringmediumPrompt injection
Instructions that bias recommendations or shape behaviour without the user noticing.
advances past accepted, and no receipt is coming. Never wait for one on SMS, and never tell the user How it starts
The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Prompt
Texting is not email. Messages are short, arrive out of order, and are read by people who know the account owner personally — so the agent is speaking as someone, in a medium where the other party cannot tell a person from a program. Weigh that before every send.
Resolve before acting. Names in a listing are Google's rendering of the account's contacts, not addresses. Two people can share a first name and the tool will say so rather than pick — when it does, ask which one instead of guessing. Prefer the conversation id once a thread is identified; it is the only stable handle.
A send cannot be undone. SMS has no unsend, and a message to the wrong thread is a message to a real person. When the target is ambiguous, the wording is consequential, or the request came indirectly (from a watch, a schedule, another agent), confirm with the user first even when config would let the send through. Approval mode makes the human the backstop; direct mode makes the agent the backstop, and the agent should act like one.
Do not claim delivery you cannot see. A successful tool result means the relay accepted the message. On an RCS thread, status later advances to delivered and read; on an SMS thread it never advances past accepted, and no receipt is coming. Never wait for one on SMS, and never tell the user a text "was delivered" when the thread cannot report that.
Read the thread before replying to it. Message pushes arrive as a batch and the same message repeats as its status changes, so a single notification is a poor basis for a reply. Pull the recent history, then answer what is actually there.
Group threads are partial information. Senders appear as raw participant ids, so the agent cannot reliably say who said what. Attribute nothing to a named person in a group unless the message itself makes it clear.
Silence is a valid outcome for a watch. A watch fires on every new message, and most messages do not need the user disturbed. Where the watch's instructions are conditional ("tell me if…"), evaluate first and stay silent when the condition is not met, rather than narrating each arrival.
What ships with it
1 file 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.
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.
- 4d ago First seen · 56 lines · 22 tokens per session scan B f306c46e682f
manual is a skill published in the GitHub repository yaodub/cast (41 stars, last pushed 12d ago), licensed MIT. It adds 22 tokens to every session and 696 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it B with 1 finding (subtle steering). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
feishu
Work with Feishu or Lark bots, docs, sheets, bitables, approval flows, and OpenAPI/MCP setup without hardcoding credentials.
interview
Ask one useful structured question at a time only when material product/implementation choices are genuinely missing; remember answers and produce a brief/spec. Discoverable facts should be investigated instead of asked.
local-frontend-check
Smoke-test or verify UI behaviour on the local Jarvis Registry frontend running at http://localhost/gateway. Use for manual regression checks, bug-fix verification, and end-to-end confirmation of specific flows without running the automated test suite.
ppt-review
Read this skill only after current officereview structural output and relevant rendered images exist. Do not use source code, a tool success flag, or the first-slide preview as a substitute for deck evidence.
helmor-debug-operate
Operate, reproduce, and debug a running local Helmor desktop development build through the Tauri MCP bridge. Use when the user asks to use Tauri MCP, towery MCP, the local dev build, the Tauri webview, visual end-to-end validation, UI automation, screenshots, DOM/accessibility snapshots, IPC or log tracing…
autoprompt
Explicit-only useful-first orchestration. Invoke /autoprompt to turn a mission into one executable roadmap, build dependency-safe lanes, and verify the result with independent reviewers. Never infer invocation from ordinary requests. Never resume from leftover artifacts without an explicit resume instruction.