AGENT SKILL

mesh-operate

Triage a Meshtastic mesh through Mesh Manager and prepare the fix to the gate. Use when the operator says the mesh is down, a tracker has vanished, nobody is getting messages, or before taking the kit somewhere new. Runs reads, asks the mesh what only the mesh can answer, and hands over the decision.

License 许可
GPL-3.0
最近更新
2026-10-24
Stars
★ 0
Forks
0

01 前言 Front Matter

name
mesh-operate
origin
mesh-manager/skills
source
MilUX Ltd
audited
1789171200000
license
GPL-3.0-or-later
category
operations
audit_sha
41ac7e0b96d8f215
maintainer
MilUX Ltd
description
Triage a Meshtastic mesh through Mesh Manager and prepare the fix to the gate. Use when the operator says the mesh is down, a tracker has vanished, nobody is getting messages, or before taking the kit somewhere new. Runs reads, asks the mesh what only the mesh can answer, and hands over the decision.
audited_with
skill-safety-audit (MilUX meta-skills)
audit_verdict
pass with cautions
product_version
1.6.1
cautions_accepted
2026-09-12, Matt Odell, MilUX Ltd

03 SKILL.md 正文

Operating the mesh: triage to the gate

Load mesh-lessons first. Then work top down; stop at the first thing that explains the symptom, and say what you did not check.

1. What kind of box this is

status carries mode. A tak-server box bridges into a TAK Server, a server box has no TAK at all, and a desktop box is a laptop with a radio in the USB port. Read it before anything else, because it decides what "working" means here and which of the checks below apply. Proposing that somebody check TAK on a box that has none wastes their time and tells them you did not look.

2. The bridge and the radio

status. In this order: radio_present, bootloader, connected, watchdog, last_activity, last_forwarded, nodes_seen. A false at the top explains everything below it; report that one line and the physical action it needs. On a server box last_forwarded is not a fault signal: there is nothing to forward to.

3. Who is on the mesh, and how well

links, which is nodes with the picture: direct_snr is the SNR of the last packet that came straight from that node, the only figure that describes its link to this radio; history is the last two hundred readings, so a link getting worse shows before it fails. Split the nodes: heard here recently, heard here but stale, database only (heard_here false). A node at hops two or three is at the edge; a tracker at five per cent battery is about to leave. Say which nodes are which, by id and name. route gives the last traceroute answer for a node, hop by hop with the SNR at each hop, towards and back.

Split off the ones that are not yours. A node carrying origin_name came over a link from another site; this radio never heard it. Report those separately and say which site, because they are not evidence about your own air and nobody can walk out and find them. mesh-join has the rest.

4. What they ride

channels and config: the primary channel, the region, the modem preset. If the operator's brief (mesh_context) names a different region or channel, that is the finding.

5. Ask the mesh, at propose or above

Only what changes nothing: traceroute to a node that should be there (the answer arrives as a route; read it with route), request_position from a tracker with no fix. Say what you are sending and why first. Watch log and messages for the answer, and wait a minute before reading silence as a fault.

6. What the fleet is, and what you may change on it

register: every node the radio knows of joined with the box's register on radio id, the operator's label and holder beside the node's own name, and managed, which is true only when a read of the device itself showed this radio's public key among its admin keys. A device that is not managed cannot be changed from here at all; the only thing to say about it is "bring it to the bench". For a managed device, node_read asks it over the air for what it holds now; node_set, node_set_region, node_channel_push and node_reboot are the writes, at act or through propose. Every write answers confirmed true only when the device's own answer matched; unconfirmed with a reason otherwise; a write over LoRa can take a minute and silence is not failure. Region, slot 0 and reboot need confirm set to the device's own id, because afterwards it may be unreachable over the air.

7. Hand over the decision

Anything that would change a device, a channel or a region is above the on-air line unless you hold act. Use propose with the exact action and arguments and a rationale in the operator's words, and say what changes if they say yes and what the way back is. A region change before travel needs two QRs: one for the destination and one for the way home; say so.

Alerts, and what acknowledging does not mean

alerts is what is open. At act you can alert_ack, and it is worth being exact about what that does: it takes a row off the open list and stops it being raised again while the condition still holds. It does not fix anything and it does not mean the condition is fine. When the condition genuinely clears the acknowledgement is forgotten, so a recurrence alerts afresh. Acknowledge when the operator has decided they know, not to tidy a list you find untidy, and say which alerts you acknowledged. An alert that came over a link belongs to the site it came from and cannot be acknowledged here.

Before the kit goes somewhere new

Read config for the region. Read the brief for where the kit is going. If they differ, the first proposal is the region, and nothing else until the operator answers.

04 同仓库更多 Skills