- Home
- Skills
- Business & Operations
- Know What Every Agent Can Reach
More screenshots
Works with the AI Agents you already use
Know What Every Agent Can Reach
One list of every AI agent and tool you run, what each can reach, and what breaks or leaks if one is compromised.
$12
Know What Every Agent Can Reach
Example session with this skill installed
Perform a risk audit on my current agent inventory. I have three MCP servers running and a cron job that logs to a private Slack channel. Check for toxic combinations and high blast radius.
- Read your context and instructions
- Compiled the know what every
I have completed the inventory audit and risk assessment. I identified 4 active identities, including one shadow agent not present in your primary manifest.
Agent Risk Register & Triage Report
| Identity | Reach/Permissions | Risk Score | Status | Primary Finding |
|---|---|---|---|---|
| mcp-sql-helper | DB (Read/Write), Slack (Send) | 75 | BLOCKED | Toxic Combo: DB Write + Slack Egress |
| shadow-proc-88 | Filesystem (Read), Network (Out) | 65 | REVIEW | Shadow agent: Unregistered local process |
| cron-logger | Slack (Send), Logs (Read) | 35 | SANCTIONED | Stale credential: Token > 180 days old |
| mcp-search | Web (Read) | 12 | SANCTIONED | Low risk: Read-only access |
Blast Radius Analysis: mcp-sql-helper
- Direct Reach: Local PostgreSQL instance (admin), Slack Workspace (webhook-user).
Lateral Path: Shares Slack webhook with cron-logger. Compromise of either provides a vector to the other's message history.
3. Worst Plausible Exit: Exfiltration of production schema or customer table data via Slack egress.
Prioritized Fix List
Revoke mcp-sql-helper: Immediately rotate Slack webhook and PostgreSQL credentials. Current risk (75) is high due to write access combined with unreviewed messaging.
2.
Identify shadow-proc-88: Confirm if this is a legitimate browser extension or a rogue agent. Defaulting to high weight until registered.
3. Rotate cron-logger token: Replace the 6-month-old Slack token to resolve the stale credential finding.
Next steps
- Manually verify
shadow-proc-88process intent and register it. - Remove write permissions from
mcp-sql-helperif only querying is required. - Export this audit pack as a PDF for your monthly security review.
Connects securely to your tools. The creator never sees your data.
About this skill
Agents multiply quietly — a coding assistant here, a workflow automation there, a browser extension, a scheduled script nobody remembers writing. Each one holds credentials and permissions, and without a list, nobody can say what each can reach, what leaks if one is compromised, or which credential to revoke first. This skill builds that list once, scores it honestly, and keeps it current.
Inside the package: the skill file covering the register format, the discovery sweep, scoring, and triage; a filled example register entry; and the risk-scoring reference with the full factor table. Each entry captures identity and owner, the agent's purpose in one sentence (its stated intent is what later drift gets measured against), every connected tool with its actual permissions — read is a different agent from read-plus-send — the data types it can touch, whether a human reviews its output, and a governance status: sanctioned, review, or blocked. A field that's unknown gets written as unknown. A register that guesses is worse than one with gaps, because the gaps are themselves findings.
A run goes like this: before scoring anything, a read-only sweep of the machine hunts for agents nobody registered — agent and MCP config files, browser extensions with page access, running agent processes, scheduled jobs and autostart entries — and anything found but unlisted enters the register as a shadow agent, the heaviest-scoring factor of all. Each agent is then scored 0 to 100 from named factors: dangerous permission pairs (reading an inbox plus sending mail with no human review is the classic), stale credentials, excessive permissions, no human review, repeat incidents. For the highest scorers, the agent walks the blast radius — everything it can reach directly, every other agent sharing those tools, every data type in that zone, and the worst way out. You get a fix list sorted heaviest first, with the projected score after each fix, and review dates set by band. The register documents and recommends; it never revokes anything itself. Revoking a real credential stays a human action, and the skill keeps it that way.
Cannot enumerate cloud-only OAuth grants or remote agents without manual input. Does not perform automated credential revocation. Process name matches require human confirmation.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago