Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
14 contributions to Clief Notes
Help please: job tracking dashboard across 3 Hermes agents, has anyone built one?
I'm looking for some help. I run a fire protection business in Sydney. My automations run on Hermes as three agents, one after another: - Vista EA, my executive agent, takes in the work and runs most of the admin. - It hands jobs to FireCAD, which does the design and drawings. - FireCAD then hands to the ARA desk, which does quotes, tenders and pricing. Each agent runs several workflows inside it. Right now they report to me on Telegram. We use the Kanban board built into Hermes, but only to pass tickets between the agents. It works for them. It isn't easy for me to read or act on. What I'm missing is job tracking that a person can use. I want one view that shows each job, which agent has it, what step it's at, and what is waiting on me. I also want to act on it from there: approve, reply, send it back. It has to work natively in Hermes, or through Hermes's own plugins or API. I don't want something that breaks every time we update Hermes. Has anyone built something like this? Did you use a Hermes plugin, put a friendlier screen over the Hermes Kanban, or build something separate that stayed stable through updates? A screenshot, a template, or a "don't do it this way" would all help.
0 likes • 13h
@David Vogel Thank you, much appreciated. Looking forward to the deeper answer.
1 like • 13h
@M Rukhshan Thank you, much appreciated. One job ID and a "waiting on me" column first makes a lot of sense.
Multi-day runs: a handoff file, or one state both runs read?
My multi-day jobs hand a file from one run to the next. One run died after finishing step three but before writing it down, so the next run opened the file, saw step three still open, and did it again. Both runs exited clean and I only noticed the duplicate much later. The handoff now carries the writing run's timestamp and the evidence that the step landed, so "not done" and "done but unrecorded" stop reading the same. It does nothing for a run that dies before it writes at all, and that file still looks complete to whoever opens it next. For work that spans days, do you pass a file between runs, or keep one state both runs read and write?
1 like • 4d
I use both, for different jobs. A handoff note tells the next session where to pick up. A separate run log gets one line every time a step runs, including steps that were skipped or blocked. The handoff says what the last session thought happened. The log says what actually ran.
1 like • 1d
@Leo Saraiva Fair point. The plan writes a line when a step starts and another when it ends. A start with no finish is the run that died. The log is add-only, one file a month, and I keep the old months, so a restart can read back as far as it needs.
State your prism
Someone says "the sky is red." Everybody disagrees. Then they show the prism they were looking through, and through that prism the sky is red. The claim was never false. It was missing its condition. That's how I now try to talk about agents: claim, condition, proof. "My agent works" is a red sky. "In a fresh chat, it recalled 9 of 10 facts I saved yesterday, here's the test and the output" (just an example) is a statement nobody can wave away. And it keeps me honest too: no evidence, not verified. State the prism you're looking through, and nobody can call the red sky wrong.
State your prism
2 likes • 4d
This is how I run my AI work. Every job is a card on a board, and the board has a definition of done that every card inherits. A card can't move to done on the AI's word. It needs the check written on the card, like the file existing or the test passing. "Done" with no proof stays where it is. Your claim, condition, proof is the same idea in one line.
2 likes • 1d
@Pascal Pollack The AI writes it, but only from a plan or a ruling I made. It is never made up. If nothing says what done means, the card says so and comes to me. Either way the final "done" is my mark, not the AI's.
Canadian RPAS Site Survey Generator
I've built a local web app that turns an RPAS mission into a CAR 901.27 preliminary site survey. The pilot enters the date and times, flight area, altitudes, aircraft, team and NOTAMs. The app fetches the weather, airspace, aerodromes and obstacles it can reach, works out a weather-aware Operational Volume, and produces a PDF and a KML. The PDF includes the classification with CAR citations and a printable on-site form. I made it because building a 901.27 survey by hand for every tasking takes a long time, and it’s easy to miss one of the eight factors. Now a survey takes minutes and cites the same way every time. The pilot still confirms everything on site and signs on paper, and the app never records a PIC determination. For ICM, the workspace is laid out so an agent only reads what the task needs. A short router file sends each task to one folder and one skill. Stable material (references, skills, policy data) sits apart from the per-survey folders under missions/. Each survey runs through numbered stages, 00_mission, 01_flight_geography and 02_preliminary_survey, and each stage’s output feeds the next. A short list of hard rules applies to every run. The app records facts only, never invents data, and never softens a BLOCKING or REVIEW result.
0 likes • 4d
This is the right shape for compliance work. I run a fire protection business, and our paperwork has the same rule. The tool can prepare everything, but a licensed person signs. "Never softens a BLOCKING result" is the line I'd steal. How does the app handle data it couldn't fetch? Does it block, or flag it for the pilot?
0 likes • 3d
@Cory Harrison Thanks Cory. If it fetches everything, the blocking rule only has to catch bad data, not missing data. That's a nice place to be. Good luck with EVLOS and BVLOS.
ICM Memory
Happy Monday, everyone. Extreme novice here. 😬 I have a question regarding memory in the ICM setup. I have never used Claude Code before, or setup anything like ICM. I'm basically the standard user who uses Claude CoWork attached to a local folder (at least when we could do that), but never actually uses the folder as the workflow. Regarding memory, how does ICM manage memories with Claude Code, or how do you manage memory? I know Claude Cowork has memory within projects, but with an ICM setup, does Claude Code have auto memory? If so, do you recommend using it, or do you use a memory.md file? I have seen so many different setup variations from different creators on YouTube , that my ADHD Swiss cheese brain is just about ready to melt, lol. I have seen memory.md in the global Claude folder, in projects, in the root folder, and in the workspace folders. I'm still trying to comprehend ICM and figure out just how to get started. But I'm worried as how to managed Claude's memory. Right now in co-work, I can jump into a chat that happened two weeks ago, and it still has memories from the chat and from the project. I'm basically just worried that I'm going to get started off on the wrong foot with Claude Code and have to fix everything later. I would rather start solid from the beginning and iterate when needed. By the way, I love reading everybody's posts, even though I only understand about 10% of what you all are saying, lol. But it's a great learning experience!!
3 likes • 4d
Don't worry about getting it perfect on day one. What worked for me is one short instructions file per project, which points to the other files instead of copying them. Anything worth remembering goes in a note in its own folder, not in one giant memory file. Small files the AI opens when needed beat one big file it has to read every time. Start with one project and add structure only when something breaks.
1 like • 3d
@Anthony D'Angelo Claude writes them. I don't type them. When something is worth keeping, I tell it "log this" and it files a short note in the right folder. It also has a standing rule to file what it learns when it fixes a mistake. My part is saying yes or no to what goes in.
1-10 of 14
Maciej Karnecki
3
38 points to level up
@maciej-karnecki-8818
Name is Mac, from AU.

Online now
Joined Oct 31, 2026
Powered by