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/markifact/markifact-mcp/safe-write-operationsnpx skills add markifact/markifact-mcp --skill safe-write-operationsgit clone --depth 1 https://github.com/markifact/markifact-mcpWhat 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 | $0.00028 | $0.00807 |
| Opus 5 | $0.00014 | $0.00404 |
| Sonnet 5 | $0.00006 | $0.00161 |
| Haiku 4.5 | $0.00003 | $0.00081 |
Grade A, and why
safe-write-operations 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 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.
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 — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Safe write operations
Calls to run_write_operation can spend money, pause campaigns, change budgets, edit live creatives, send emails, or charge customers. Treat every one as a production change.
When this protocol applies
The trigger is the requires_approval: true flag on the operation in the find_operations response. If requires_approval: true, you must use run_write_operation and must follow the four-step protocol below. If requires_approval: false, use run_operation and skip this protocol — no confirmation needed.
The four-step write protocol
Always follow this exact sequence before calling run_write_operation:
- State the change in plain English. Include the platform, account, object name/ID, and what will change. Example:
"I'm about to pause the Google Ads campaign 'Brand — US Search' (ID
1234567890) in account Acme US (MCC 999-888-7777). This will stop spend immediately." - State the blast radius. What spend, traffic, revenue, or audience is affected? Pull a quick number from
run_operationif you don't know. - Ask for explicit confirmation. Wait for a yes. "Looks good", "go ahead", or "do it" all count. Anything ambiguous = ask again.
- Execute, then verify. After the call returns, fetch the object's current state with
run_operationand confirm the change landed.
Hard rules
- Never batch more than 5 write operations without re-confirming. If a workflow needs 20 pauses, do the first 5, summarise, ask "continue with the next batch?".
- Never delete unless the user said "delete" (not "remove", not "pause"). Prefer pause/disable over delete every time.
- Never change budgets by more than 50% in a single call without an extra confirmation step. Big budget moves are easy to fat-finger.
- Never run a write operation immediately after a model-invoked discovery. The user must have asked for the change in their own words first.
- Never assume a connection. If the user has multiple Google Ads accounts and didn't specify, ask.
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 · 44 lines · 28 tokens per session scan A 0dcc4de10cdf
safe-write-operations is a skill published in the GitHub repository markifact/markifact-mcp (48 stars, last pushed 14d ago), licensed MIT. It adds 28 tokens to every session and 807 once invoked, about $0.0001 per session on Opus 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 2026-08-30.
Other skills, from other repositories
hwp
Use this skill whenever the user wants to read, create, or edit Korean Hangul Word Processor documents (.hwp or .hwpx files). Triggers include any mention of 'hwp', 'hwpx', '한글 문서', '아래한글', '한컴오피스', or uploading/attaching .hwp/.hwpx files. Also use when extracting text from Korean reports or producing Korean-formatted…
dev
Software development planning. Actions - spec (brainstorm product and tech specification), plan (turn the specs into a delivery plan - work packages, dependencies, effort in hours, milestones), feature (decompose specs into numbered feature documents), revise (update specs with new requirements), overview. Planning…
ws
Create and run workstations - permanent areas for routine, repeated work (independent of projects). Actions - new WS, new subWS, new area, new routine, run, from project, list, overview. Hierarchy depth (2 or 3 levels) is fixed at creation.
project
Create and manage the project structure - projects, sub-projects, and work packages (WPs). Actions - new project, new subproject, new WP, overview. Hierarchy depth (2 or 3 levels) is fixed at creation. All work happens in WPs; projects and sub-projects are pure structure.
mcp-playbook
Deep reference / playbook for correctly driving the crevideo-reach MCP — TikTok Shop affiliate (分销) work: tool map (79 tools), the create flow, TC automation-wrapper vs direct entity, Open Collaboration (OC), read-side tool choice, filter schema, ID-space gotchas, message components, naming, reporting gotchas, sample…
creator-fit-scoring
Scores one or more TikTok Shop creators for real sell-through fit using evidence (conversion over followers), returning a fit verdict + recommendation strength + outreach suggestion, with a lookalike-expansion recruiting mode. Use when the user wants to score, rank, evaluate, shortlist, compare, or find lookalike…