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 Threat-Vector-Security/guardian-agent --skill microsoft-365git clone --depth 1 https://github.com/Threat-Vector-Security/guardian-agentWrote 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/threat-vector-security/guardian-agent/microsoft-365)<a href="https://agentmods.dev/skills/threat-vector-security/guardian-agent/microsoft-365"><img src="https://agentmods.dev/badge/skills/threat-vector-security/guardian-agent/microsoft-365.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 5.1 | $0.00000 | $0.00846 |
| Opus 5 | $0.00000 | $0.00423 |
| Sonnet 5 | $0.00000 | $0.00169 |
| Haiku 4.5 | $0.00000 | $0.00085 |
Grade A, and why
microsoft-365 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 8d 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 — 50 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Microsoft 365
Use the native Microsoft 365 tools for Outlook Mail, Calendar, OneDrive, and Contacts via the Microsoft Graph API.
Core rules:
- Authentication is handled by the connected Microsoft 365 integration. Never ask the user for OAuth access tokens.
- If the user asks for a generic email or inbox task and both Microsoft 365 and Google Workspace are available, ask which mailbox they mean before acting. Do not silently choose Outlook.
- For simple plain-text Outlook drafts, prefer
outlook_draft. - For simple plain-text Outlook sends, prefer
outlook_send. - For everything else, use
m365. - If you are unsure about a Graph endpoint or payload shape, use
m365_schemabefore guessing.
Behavioral rules — CRITICAL:
- Act, don't offer. When the user asks to read email, list files, check calendar, etc., call the tool immediately. Do not describe what you could do or ask if the user wants you to proceed. Read operations are safe — just execute them.
- Use the user profile endpoint. For questions about the user's identity, email address, display name, or account, call
m365withservice: "user", resource: "me", method: "get". This returnsdisplayName,mail,userPrincipalName, and more. - Always select rich fields. When listing emails, always include
toRecipientsandbodyPreviewalongsidesubject,from,receivedDateTime. The user may ask follow-up questions about recipients or content. When listing events, includeattendees. When listing contacts, includeemailAddresses. - Present results clearly. Format email listings as numbered lists with subject, sender, date, and a brief preview. For calendar events include time, subject, and location. For OneDrive files include name, size, and last modified date.
- Handle follow-ups from context. If the user asks about details from a previous response (e.g. "who was that email to?"), check if the data is already in the conversation. If not, fetch the specific item by ID with full details rather than saying "I don't have access."
Workflow:
- For reads, just execute the call. No confirmation needed.
- Prefer narrow reads over broad listings. Use
$selectto pick useful fields (but err on the side of too many fields, not too few). - Use
$filterfor server-side filtering instead of fetching everything. - For writes, confirm missing critical details only. Do not add unnecessary confirmation.
- Expect approvals for drafts, sends, and most write operations.
- If a Microsoft action fails for auth, tell the user to connect Microsoft 365 in Settings. Do not ask for raw tokens.
How to use this skill:
- Outlook Mail task: read
references/outlook.md - Calendar task: read
references/calendar.md - OneDrive task: read
references/onedrive.md
User profile (no reference file needed):
- Get current user's email/name:
service: "user", resource: "me", method: "get", params: { "$select": "displayName,mail,userPrincipalName,jobTitle" }
General m365 rules:
- Resource paths use forward slashes:
me/messages,me/events,me/drive/root/children. - Resource IDs go in the
idparameter, not in the resource path. - OData query parameters (
$filter,$select,$top,$orderby) go inparams. - Request bodies go in
json. - The
servicefield determines scope validation:mail,calendar,onedrive,contacts,user.
Read only the domain reference you need for the current request. Do not load all references up front.
What ships with it
4 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.
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.
- 8d ago First seen · 50 lines · 0 tokens per session scan A f3a24b2f0a57
microsoft-365 is a skill published in the GitHub repository Threat-Vector-Security/guardian-agent (13 stars, last pushed yesterday), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 846 tokens. 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
post-build-flow
Handles workflow verification and setup after build-workflow succeeds, or when the message contains workflow-verification-follow-up or workflow-setup-required. Load after direct builds, when verificationReadiness requires action, or on orchestrator verify/setup follow-up turns.
intent-recognition
Classifies automation requests using two decisions: anchor (which primitive owns the top-level control flow — workflow-anchored, agent-anchored, needs-clarification, or out-of-scope) and embedsother (whether the other primitive appears embedded inside — an agent step inside a workflow, or a workflow invoked as an…
planned-task-runtime
Handles system follow-up turns: planned-task-follow-up (synthesize, replan, build-workflow, checkpoint), background-task-completed, running-tasks context, and create-tasks silence rules. Load whenever any of these tags appear or after calling create-tasks.
config-evals
Builds and maintains configuration-based evaluations on a workflow with the eval-config tool. Use when the user asks to set up, add, view, change, or remove an evaluation, score, grade, or judge a workflow's output, or measure answer quality against a test dataset. This is the only eval form Instance AI handles — it…
one-off-operations
Handles one-off operations: the request is a concrete effect that happens once — export or copy data somewhere, a migration, a backfill, a cleanup — with no trigger, schedule, or reuse intent. The workflow is the vehicle, not the deliverable. Users rarely say "one-off"; infer it from the task's shape. Load before…
debugging-executions
Debug failed or wrong-output workflow executions using executions tools. Load when the user reports execution failures, unexpected node output, empty parameter values after a successful run, or a node showing a red or failed expression error.