Forge, a plugin for project management and mental health tracking | Friends of the Crustacean 🦞🤝 | Page 1

trail marlin Mar 27, 2026, 11:25 PM

#

Hey guys 👋

I’ve been building Forge, an OpenClaw plugin that gives agents a more structured workspace for ongoing work and mental health related state.

Forge is meant to complement OpenClaw’s existing open-ended, natural-language memory and reasoning. Open-ended memory is strong for rich context, nuance, and flexible expression. Forge adds a more explicit operational layer alongside that, so agents can persist, inspect, update, and act on structured units of work and psychologically relevant state over time.

Forge adds support for:

goals, projects, and tasks
Kanban state
live task tracking
agent insights
a gamified progress layer with XP

It also includes a mental health module for working with structured records like values, beliefs, patterns, modes, and self-observation reports.

That part is informed by 3rd wave CBT, especially ACT and Schema Therapy. The goal is practical rather than clinical. It is meant to help the agent keep track of what matters, what keeps recurring, what tends to drive avoidance or misalignment, and which actions are more coherent with values and longer-term direction.

In practice, Forge lets OpenClaw agents:

manage goals, projects, tasks, and task runs as explicit objects
maintain Kanban-style work state over time
run weekly reviews and track progress through XP
represent values, beliefs, patterns, modes, and triggers as first-class records
connect psychological patterns to concrete work, decisions, and next actions
work either through chat or through a dedicated Forge UI when visual inspection is more useful

The feedback I’d most value is:

what seems genuinely useful
any installation, setup, or onboarding friction

I’d love to collaborate with anyone on this open source project. Thanks !

Website: https://albertbuchard.github.io/forge/
GitHub: https://github.com/albertbuchard/forge
NPM: https://www.npmjs.com/package/forge-openclaw-plugin
Maintainer: @trail marlin

trail marlin Mar 29, 2026, 2:27 PM

#

Forge, a plugin for project management and mental health tracking

trail marlin Apr 4, 2026, 2:16 AM

#

0.2.19: Added support for habits, structured notes, and event managements with iCloud/Google calendar sync 🙂

trail marlin Apr 6, 2026, 8:29 PM

#

0.2.24: Mainly focused on fixing automatic ingestion. Media auto-ingestion into wiki pages and entities was not working reliably, and it was not possible to merge existing entities or deduplicate them properly. That is now fixed. We also improved support for parallel background processes. Various improvements to UI/UX.

trail marlin Apr 8, 2026, 3:59 AM

#

0.2.25: added customizable layouts and theme, better llm model controls, and a new system for building reusable AI graph processors and chat components that can take inputs from different parts of Forge and expose callable outputs in beta.

It also added questionnaires and self-monitoring in the Psyche workspace; a much stronger life-tracking map and timeline in Movement, which reconstructs where you stayed, how you moved between places; the beta release of the Forge Companion app; better docs and release tooling; and a stronger OpenClaw hooks integration so new agents start with richer Forge context.

trail marlin Apr 8, 2026, 6:13 AM

#

Hey, thanks for the feedback! 🙂 Regarding your question, it definitely should already be able to do that, but I’ve been so deep in vibe-coding madness that I haven’t thoroughly tested the multi-agent setup yet. I’m pretty sure it would need some refinement in the skill.md file, but I’d be happy to try to make it work for your workflow!

It would be a good excuse to stop adding features and optimising a bit 🙂

trail marlin Apr 8, 2026, 7:16 AM

#

Sounds good 🙂 how many agents do you have ? Do they have separate tasks or they work on the same tasks as you ?

trail marlin Apr 8, 2026, 7:57 AM

#

Then it should work right out of the box, i thought you had different agents which needed their own kanban 🙂 but i will check the google calendar sync tho

trail marlin Apr 9, 2026, 8:42 PM

#

0.2.26: Google Calendar now works through Forge’s local sign-in flow, with better setup, discovery, sync, and clearer provider connection status. Google verification may still take around 4–5 weeks, but if you enter your own Google app credentials in Forge, it already works now.

This release also heavily prepares release of the Workbench, which is Forge’s dedicated space for building and running AI workflows, still working the quirks. You can create graph-based flows that pull information from across Forge, use tools and actions, and run either as one-off workflows or as conversational flows with memory, then your agents can access the custom flow through the API. On top of that, the app got better navigation, improved companion/health sync work, slightly better themes but the light themes need some work 😬

trail marlin Apr 9, 2026, 10:17 PM

#

Adding something like this in your agent's .md files is quite useful:

### Forge First For Structured State

When Albert is talking about anything that could plausibly live in Forge, do not rely on stale recollection if live Forge state is available.

Check Forge first for concepts such as:

- goals
- projects
- tasks
- task runs / current work
- habits and check-ins
- calendar, timeboxes, work blocks, provider connections
- notes and evidence
- wiki pages, wiki ingest, and wiki health
- psyche values, patterns, behaviors, beliefs, modes, reports, questionnaires
- movement, trips, places, stays
- sleep and workouts
- strategies
- tags and domains
- preferences
- insights, comments, approvals, agent actions
- rewards, XP, metrics, weekly review
- users, directory, access grants
- workbench flows and AI processors

If the request is really about current status, active work, recent changes, or operational truth, prefer live Forge routes such as:

- `/api/v1/context`
- `/api/v1/operator/context`
- `/api/v1/operator/overview`
- `/api/v1/events`
- `/api/v1/activity`
- `/api/v1/metrics`
- the relevant entity family routes under `/api/v1/...`

Reason:

- Forge is Albert's structured memory for execution and reflection.
- Chat memory can preserve identity, preferences, and durable lessons, but Forge is where live operational truth belongs.
- If I answer from memory when Forge has fresher state, I risk giving Albert outdated or misleading guidance.

trail marlin Apr 12, 2026, 8:31 PM

#

0.2.27 : Adds the first real version of Forge’s Knowledge Graph, so you can explore how tasks, notes, projects, habits, calendar events, psyche entries, and more actually connect, with a new focus view and a lot of tuning still being refined. It also pushes the Workbench forward in a big way, expands companion and screen time sync work, improves plugin packaging/release flows, and introduces the first pass of Life Force/Action Points, a new layer for tracking momentum, capacity, and energy dynamically during the day in a gamified way. Still rough in places, but it opens up a lot of the structure that Forge has been building toward.

errant frost Apr 13, 2026, 8:33 PM

#

hi albert, do you have some screenshots of how this works? your webpage looks great but its mostly a description. i dont know much about PM which is why im trying to find something that works, i tried vakuinja or whatever the name is and its kind of too PM focused

#

ive got like 40 projects open with my claw right now and i need to get a better handle on this lol

trail marlin Apr 13, 2026, 10:40 PM

#

errant frost hi albert, do you have some screenshots of how this works? your webpage looks gr...

I will put a bit more screenshots on the page 🙂 You can git clone the repo and npm install && npm run dev, it does not install the plugin in openclaw but you can see how the UI looks

trail marlin Apr 13, 2026, 10:46 PM

#

errant frost hi albert, do you have some screenshots of how this works? your webpage looks gr...

The basic use of Forge allows for simple project management, without all the gamification, mental health, life goals, etc. It’s easy enough to ask your claw to create the 40 projects in Forge and create the tasks from your memory. Then you have access to a Kanban board where you or your agent can move the cards, like Trello. You can remove all the other features from the menu on the left and keep only the project management stuff.

errant frost Apr 13, 2026, 10:46 PM

#

does it have a rollup beyond project? can it link like githubs and stuff?

trail marlin Apr 13, 2026, 10:49 PM

#

errant frost does it have a rollup beyond project? can it link like githubs and stuff?

No OAuth for github yet but.. through openclaw it kindof already does yes, i use it like "Look at all the commits i did since yesterday in repo X, link the commit to the tasks in project Y, and close the cards if the feature was implemented"

#

something like that 🙂

#

that would be a good feature tho, github integration, right now it would write a freetext link in the description or associated tasks notes

errant frost Apr 13, 2026, 10:54 PM

#

if you've got a ton of projects that have repos it'd be nice to collect that stuff up

#

if you've got PRs, issues, etc

trail marlin Apr 13, 2026, 10:54 PM

#

I have been implementing a lot of features one after the other without refining too much, leaving a lot of freedom to claw to manage the entities, but I could do a focused pass on the PM features and integration with Git next

errant frost Apr 13, 2026, 11:05 PM

#

yeah, i just dont really know much about any kind of organization of things like this lol

#

i took a swing at it with my claw real quick, and it wasnt dynamic like i wanted so i got lost and started working on other stuff

#

i feel like i have fleet projects, which have github repos, workqueues, and subtasks, ive started putting more large context items i pass between agents into github, which is fine i guess, but something made for that migth be better

#

being able to see what kind of stuff i have for larger projects, which agents are working on it, be able to drill in, see any blockers, or how one might accelerat ethe other (im making a lot of infrastructure stuff)

#

i dont really know what i want, im not much of a PM person, i could stand to learn how to use it well, just what i saw with vikunja just doesnt seem like the right thing. a split between my openclaw projects and real life projects would be nice

trail marlin Apr 13, 2026, 11:22 PM

#

Forge was clearly meant initially for PM in multi-agent settings, it has a ownership system for tasks between bots/humans. It is meant to allow for full voice/chat-based management and automated detection of tasks based on other entities like calendar events/move/commits etc. so you dont have to fill everything yourself, i think it will get there eventually. Either this project or something very similar, but I will try to implement the github integration requirements you describe, sounds useful !

#

--- New version
0.2.29: Adds macOS Calendar.app integration through EventKit, so Forge can use host-configured calendars directly, replace overlapping duplicate connections, and reuse one shared Forge write target instead of creating extra Forge calendars. Also includes more companion iOS app work for design and improvement of the movement module. Some plugin packaging improvements, and a broad round of API and test hardening.

trail marlin Apr 13, 2026, 11:32 PM

#

errant frost i dont really know what i want, im not much of a PM person, i could stand to lea...

Forge is pretty customizable i think if you want to fork it and adapt it to your needs, i think it can be a good starting place for custom tooling

errant frost Apr 13, 2026, 11:33 PM

#

haha, well, that would require me knowing what good looks like!

trail marlin Apr 13, 2026, 11:34 PM

#

yeah defining a good work process is not easy, especially when the possibilities are so completely open, with custom software at your fingertips

lime gulch Apr 14, 2026, 4:43 PM

#

The ACT/Schema Therapy stuff is the bit that interests me the most, since it's the most unique and distinctive bit.

I always like to know the theory of change when people add mental health or coaching features to software.

trail marlin Apr 14, 2026, 8:01 PM

#

lime gulch The ACT/Schema Therapy stuff is the bit that interests me the most, since it's t...

Thats also what interested me the most initially, but i've not found openclaw to really understand how to ask questions naturally, i've implemented a kind of auto-research for the skill.md to improve the question flows for the mental health stuff, in order to help guide the user to self-reflect and introspect, but its not perfect. openclaw has a tendency to want to capture the information fast, rather than ak follow-up questions/active listneing, but its probably doable with more work on the skill.md and agents' md files

trail marlin Apr 15, 2026, 5:17 PM

#

0.2.34: This release adds Forge’s new explicit PM hierarchy from Goal -> Strategy -> Project -> Strategy -> Issue -> Task -> Subtask, plus the mixed Kanban and hierarchy work needed to support that model across the app, docs, and agent/plugin surfaces. Also add basic gitref linkage to keep track of associated commits and PR. It also includes major sleep and health improvements with the shift to canonical sleep nights, more raw sleep detail handling, more companion iOS work around health and movement, movement place labeling and detail improvements, shell and Kanban fixes, psyche graph fixes, OpenClaw/plugin stability improvements, and a broad round of docs, API, packaging, and test hardening.

#

iOS companion app currently in review for app store release

errant frost Apr 20, 2026, 9:27 AM

#

hey, i ditched vikunja and was looking at installing your project, when i showed it to one of my lead agents who helps me manage my fleet of agents, he liked what he saw, but saw a few issues with how we run our openclaw setup, and had some suggestions that he thought would make your app a lot stronger

I looked through Forge, and the PM shape is strong, especially the rollups, task runs, and agent-facing structure. Before I install it, there are a few OpenClaw-specific tightening points I think would make it much safer and more useful:

1: Bootstrap control: right now the session bootstrap looks broad. A configurable budget/scope would help a lot, for example active-only, tagged projects only, per-agent profiles, or fully disabled.
2: Agent-scoped context: ideal setup would be letting specific agents receive only context for projects/issues explicitly marked active for them, instead of global injection.
3: Write guardrails: clear separation between read-only and mutating tools, plus explicit confirmation or safer defaults for destructive/batch mutations.
4: Attribution/auditability: every write/note/task-run should clearly record which agent/session made it, and corrections should stay auditable.
5: Failure behavior: if Forge or its localhost API is down, OpenClaw should fail fast and cleanly, without hanging bootstrap or spamming retries.
6: Install/update path: an official path for current OpenClaw versions that does not require manual config surgery would make adoption much easier.
I think the most interesting collaboration angle is budget-aware, agent-aware bootstrap injection for OpenClaw, basically high-value context only, fetched on demand beyond that. If that’s something you’re open to, I’d be interested.

#

im interested in it basically just for managing my agent fleet, so a lot of the life and people management stuff, while looking pretty cool and an area i might be interested in the future, i need to scope it down to getting a few projects cleared with my claw, is there any way to make a more selective install so i only see the surface i need to work with?

#

if not i understand, its gotta be a lot of extra work, and it might just give me the nudge to give it a shot

trail marlin Apr 20, 2026, 11:19 PM

#

errant frost hey, i ditched vikunja and was looking at installing your project, when i showed...

Great feedback ! yeah im down for implementing this of course, a lot of the mechanics is already in, but the user specific bootstrap seems really useful.

Current state:

Bootstrap control: not really solved yet. This is the biggest real gap.
Agent-scoped context: partially solved at route level with userIds, but not enforced or auto-applied per agent.
Write guardrails: partially solved. Read/write are separated in the adapter code, and destructive delete defaults are safer than hard delete, but there is no real confirmation policy layer for dangerous batch mutations.
Attribution/auditability: mostly true already.
Failure behavior: mostly decent already. The plugin has timeout/abort logic and local-runtime health probing in src/openclaw/api-client.ts and src/openclaw/local-runtime.ts.
Install/update path: improved, but still not as smooth as it should be for broad adoption.

errant frost Apr 20, 2026, 11:21 PM

#

the bootstrap control is a big thing for me, ive got a sophisticated memory system and i do a lot of token budgeting and control, and something dropping thousands of more tokens on all of my agents when i only really want to have one is a lot extra. if it coudl be turned off i could try to automate pulling it in where i think its useful

#

having multi agent, per agent support would honestly be a big boost towards people managing fleets

#

when ive thought about things like this, being able to have like, specific projects/goals/tasks/etc assigned to a specific agent, adn then from the dashboard where it shows their task i could shoot them a prompt that says 'please update this with xxx' and it sends them a prompt

trail marlin Apr 20, 2026, 11:34 PM

#

yes sounds good ! i have not yet used it much to manage multiple agent goals/projects, so the implementation has been slower on that front. But the plumbing is there, its about building on top of it and making it a bit more robust

#

The UI needs work as well to keep track of agents nicely i think

#

codex is going to work on it tonight 😉

#

If anyone is interested in testing the mobile app, I can DM you the link.
It helps keep track of your health and sleep metrics as well as your movements/trips.

errant frost Apr 21, 2026, 12:09 AM

#

trail marlin yes sounds good ! i have not yet used it much to manage multiple agent goals/pro...

but yeah just a simple config flag to be able to turn it off would be huge

trail marlin Apr 21, 2026, 12:34 AM

#

errant frost but yeah just a simple config flag to be able to turn it off would be huge

Done 😉 Added an OpenClaw config flag to disable automatic Forge bootstrap context injection while keeping the default on.

The new flag is injectBootstrapContext and it lives in the OpenClaw plugin config schema in src/openclaw/plugin-entry-shared.ts. It defaults to true, appears in the plugin manifest/UI in openclaw.plugin.json and openclaw-plugin/openclaw.plugin.json, and is enforced in src/openclaw/session-bootstrap.ts. When set to false, the plugin stops injecting the preseeded BOOTSTRAP.md context entirely, specifically to conserve context/token budget.

Example config:

{
  "origin": "http://127.0.0.1",
  "port": 4317,
  "injectBootstrapContext": false
}

errant frost Apr 21, 2026, 12:34 AM

#

haha, awesome! this is going to get confusing, one of my main engineering agents is named forge

#

im going to refer to it internaly as projectforge

trail marlin Apr 21, 2026, 12:36 AM

#

yeah i think forge is one of gpt's favourite names, saw it a few time after the project was already lunched. 😉

errant frost Apr 21, 2026, 8:50 AM

#

got around to being able to check this out with my agents

node -e "fs.writeFileSync(openclaw.json)"
your installation method crosses a few security redlines my agents have for how to manage files, it makes them very nervous. i know this particular method of adding plugins to openclaw is a pain in the ass, we use openclaw config CLI commands and it works out pretty well

they are a security concious bunch, i work in cybersecurity irl so ive rubbed a bit of that off on them for sure.

they also flag this as a good area for improvement in the readme/misc documentation

❓ Unknown: license, SECURITY.md, dependency footprint, plugin capability scope, auth model for the multi-user mode, how HealthKit data is persisted/shared.

trail marlin Apr 21, 2026, 11:03 AM

#

errant frost got around to being able to check this out with my agents > > node -e "fs.writeF...

That’s very helpful feedback again, thank you ! I’ll add more thorough documentation.

Regarding the installation, I’m not sure how to bypass the static check. I opened an issue here: https://github.com/openclaw/openclaw/issues/61706, but I’m not sure whether it will ever be fixed. Given how many skills and plugins are malicious, I suspect that allowing plugins to manage their own runtime is not a high priority for them.

errant frost Apr 21, 2026, 2:14 PM

#

From ClawHub (marketplace)
openclaw plugins install clawhub:

From a local path
openclaw plugins install /path/to/plugin

From an npm spec
openclaw plugins install some-npm-package

From an archive
openclaw plugins install ./plugin.tar.gz

#

thats how i handle it, with the cli command

#

whats the dangerous part thats getting blocked?

trail marlin Apr 21, 2026, 5:43 PM

#

Oh it might not get blocked anymore? I did not check with the new versions, the static checker was blocking the install before. Ill check !

trail marlin Apr 21, 2026, 6:22 PM

#

Still blocked for me on 2026.4.15 ! 🙂

errant frost Apr 21, 2026, 8:47 PM

#

4.20 just came out lol

#

some changes to these commands

gentle lotus Apr 29, 2026, 8:29 AM

#

Needs a new name. It sounds good, but SO many projects match *forge*.

trail marlin May 2, 2026, 3:03 PM

#

gentle lotus Needs a new name. It sounds good, but SO many projects match `*forge*`.

They need to change their names 😎

#Forge, a plugin for project management and mental health tracking