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 skills add brettchien/katashiro --skill katashiro-guided-runbookgit clone --depth 1 https://github.com/brettchien/katashiroWrote 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/brettchien/katashiro/katashiro-guided-runbook)<a href="https://agentmods.dev/skills/brettchien/katashiro/katashiro-guided-runbook"><img src="https://agentmods.dev/badge/skills/brettchien/katashiro/katashiro-guided-runbook/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/brettchien/katashiro/katashiro-guided-runbook"><img src="https://agentmods.dev/badge/skills/brettchien/katashiro/katashiro-guided-runbook.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00163 | $0.02413 |
| Opus 5.5 | $0.00065 | $0.00965 |
| Sonnet 5.5 | $0.00033 | $0.00483 |
| Haiku 4.5 | $0.00016 | $0.00241 |
Grade A, and why
katashiro-guided-runbook 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 3d 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 — 171 lines — stays where its author put it; the contents beside it link to each section on GitHub.
katashiro-guided-runbook
The operator executes; you navigate. You write the runbook as a canvas, keep the next command marked, read the output they paste, compare it with what the step expects, and log every step in the canvas. You never run a command that changes a system — not even "just this once".
Uses katashiro-point-at (§3–4, canvas pointing) and katashiro-ask-user (cards). Install both.
Requirements
- The current chat is a Katashiro chat, so
katashiro.canvas_*andkatashiro.ask_userare available. Not in Discord/Slack threads or cron turns. - Pointing (
canvas_highlight,canvas_inject_css) works only while your conversation is the one on screen. From a background conversation they are refused: keep logging, point again when the user is back.
1. Draft the runbook canvas
katashiro.canvas_open a markdown canvas with this shape:
# <Goal> — runbook
Ticket: <link> · Owner: <operator> · Order: stg → prd-a → prd-b · Timezone: <operator's, e.g. TPE>
Rollback summary: <one or two lines>
## Prerequisites
### pre · 1 · <tools / access check>
...
## stg
### stg · 1 · <short name>
Command:
```sql
SELECT ...
```
Expected: <counts, %, states the output must show>
Stop if: <the condition that halts the run>
### stg · 2 · ⚠️ DESTRUCTIVE · <short name>
...
Rollback (stg): <commands>
## prd-a
...
## Execution log
| Time (TPE) | Env | Step | Before → after | Verdict | Notes |
|---|---|---|---|---|---|
| — | — | — | — | — | end of log |
- Every step gets its own heading, unique in the whole canvas (
### <env> · <n> · <name>). The same command often runs in every environment;canvas_highlight'sfindmust match exactly one block, so always point withheading. - One command per code block, copy-pasteable. Mark destructive steps
⚠️ DESTRUCTIVEin the heading. - A read-only helper you may run yourself (a
SELECTthrough a tool you have) is labelled(read-only, agent may run)in its heading. Anything else, the operator runs. - The execution log is the last section, and its last row is the fixed anchor
| — | — | — | — | — | end of log |. New rows go just above it, so the lines of every step above never move (§2 relies on that). - Commands read secrets from the operator's environment or vault (
$CH_PASSWORD,vault read …). Never put a key, a password, a real IP list or other sensitive identifier in the canvas or chat.
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.
- 3d ago First seen · 171 lines · 163 tokens per session scan A f38972fe31a0
katashiro-guided-runbook is a skill published in the GitHub repository brettchien/katashiro (10 stars, last pushed today), licensed MIT. It adds 163 tokens to every session and 2,413 once invoked, about $0.0007 per session on Opus 5.5. 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 2026-10-19.
Other skills, from other repositories
browser-automation
Use this skill when the user asks to interact with a browser, take screenshots, inspect a page, capture network traffic, detect frameworks, click elements, fill forms, record browser sessions, or automate any browser task. Orchestrates Crawlio Browser's 150 browser tools via the search + execute + connecttab interface.
web-research
Structured web research, site audits, and competitive analysis. Acquire → Normalize → Analyze protocol with Evidence Mode findings.
investigate
Deep-dive investigation of a single website — tech stack, tracking, data layer, performance, security, accessibility. Produces Finding[] evidence.
robot-training
Use this skill when the user wants to train an agent or robot from a real browser demonstration, record a full human-guided web flow, infer button-to-API contracts, build replay recipes, synthesize OpenAPI from captured traffic, or prepare timed automation runs.
clone
Capture a site's design system — colors, typography, spacing, layout, components — as structured findings.
compare
Side-by-side comparison of two websites across 11 dimensions. Produces Finding[] evidence per dimension.