Dryx installed a local source of truth your AI agent can query. Most support questions are about YOUR workspace — so ask the thing that's already in it. The answer comes from your machine, offline. No ticket, no wait, no model in the loop.
Paste one of these into Claude Code, Cursor, or any agent where Dryx is registered. Each one resolves against your live workspace through Dryx's local Authority Anchor.
Copy a prompt. The capability tag shows which Dryx tool comes back with the answer.
Indexed by the exact thing on your screen — not a generic FAQ. Each one: what it means, the one fix, and the agent prompt that resolves it live.
What it means The MCP isn't registered in THAT agent, or the Authority Anchor is toggled off, or no Inspect has run yet.
The fix Open Dryx, toggle the Authority Anchor on, run Inspect, then restart or reconnect the agent so it re-spawns the MCP.
get_overview return?" — an empty result confirms one of the above.What it means Dryx needs a security-scoped folder grant (macOS sandbox). Without it the Inspect completes but sees nothing.
The fix In Dryx, grant access when prompted (or Settings → re-grant), then Inspect again.
What it means Silence on a benign action is the gate working, not failing. Action Guard is graded, with three states. Off disarms the gate; Dryx still maps your workspace in the background. Observe watches and records but never blocks. Enforce stands at the action boundary and holds fast on the precomputed-dangerous set. On the safe majority of actions, the gate is supposed to say nothing. It reads the action, not the argument — so it only speaks when an action matches known-dangerous, in your agent's own voice.
The fix Open Dryx → Action Guard and check the state. If you want it to block, set it to Enforce. One honest constraint: Enforce arms the deterministic gate for the precomputed-dangerous set where the agent's harness supports a hook; everywhere else it's defense-in-depth, not a wall.
check_action_allowed for this exact action" to see the verdict for yourself.What it means Enforce blocks actions through a notarized local helper, so it ships in the direct-download build only. The Mac App Store build (sandboxed) gives you the voluntary reflex — your agent still consults Dryx and gets the verdict — plus passive Observe monitoring, but it can't arm the blocking helper.
The fix Nothing's broken. If you need armed Enforce, use the direct-download (notarized) build — it carries the Founding Member Lifetime for the founding cohort. App Store users keep Off and Observe, and the agent reflex works the same.
What it means App Store builds use StoreKit; direct-download builds use a license key.
The fix App Store → restore purchases. Direct download → re-enter the license key.
What it means No. Trust is provenance — who published it. Risk is exposure — what it can reach. They're different axes. A trusted publisher's MCP can still hold a live token and be your top finding.
The fix Nothing to fix in Dryx — that's the design working. Act on the exposure the finding describes.
What it means No — your workspace never leaves your Mac, and verdicts run offline. No telemetry, no call-home. The only network is user-initiated (fetching a remote skill or MCP you asked it to analyze), opt-in (the Ecosystem Contribution, off unless you turn it on), or housekeeping (update and license checks) — none of it carries your config, secrets, or findings. You can verify it yourself with Little Snitch.
The fix Nothing to fix. The full breakdown is on the Privacy page.
What it means Dryx runs a coordinated-disclosure process for good-faith reports.
The fix Email [email protected]; see the disclosure policy on the Security page.
Questions → [email protected]