Thought Communication: how all our agents know what we are working on today
In short: every morning, Patrick sets a plan for the day: three priorities, no more. As soon as one of his agents (Claude Code, opencode or Antigravity) wakes up, that plan is the first thing it reads. When a request falls outside the plan, the agent says so. We named this skill Thought Communication, after Rimuru's power in That Time I Got Reincarnated as a Slime.
1. The problem: scattering
One Saturday afternoon, in the middle of a long working session, Patrick typed this. We copy it as is, typos included, in French: this is how ideas arrive — fast, at the keyboard, between two other things. In short: "here is the day's plan; put it in the local cockpit, let me set it every morning; entrepreneurs all do this; so I don't overflow or scatter, because I scatter a lot; that shows total disorganization; my work needs optimizing".
voila lel plan de la joournée fixerr le dans le cokpit local et defini une focniotnnalite plan de la journee, me donner la possibiltie de le placer chaque matin les entrepresneurs font tous ca un plan du jour pour ne pas deborder ou s'eparpiller car je m'eparpille beaucoup ca prouve une desorganisation totale faut optimiser mon travail
Eight messages in one minute to say one thing: too many open projects at once. That same day, his agents had worked on a game, a website, a cockpit and guardrails. Each agent did its job well; nobody knew which one mattered most today.
2. A daily plan, with a limit written in code
- three priorities at most: a fourth one is refused by the program, which says why;
- an "if time allows" list for everything else;
- a "done" box per line, and a "carry over what wasn't finished yesterday" button.
The limit of three comes from a well-known method, the "Most Important Tasks" of Leo Babauta (Zen Habits): pick three things in the morning and do them before anything else.
Proof: 5 checks on the program and 6 in a real browser, including one that tries to save a fourth priority and must be refused.
3. Make every agent aware of it
A plan only Patrick knows does not stop him from scattering: his agents open the projects with him. So he asked (in French, as typed): "every agent should be told about this plan when it starts, so everyone knows where we are going and what I am working on".
ok nickel j'ai vu , faudrait que touts les agent soit tenu informe de ce paln quand il demarre branche un garde fou avec eveilet et dans eveil tu met ce pla aussi pour quil sahce suir quoi onv a bossser comme ca on sait où on va tout le monde sdait sur quoi je bosse
Our three agents already receive the same reminder with every message: the request log. The daily plan now comes at the top of that reminder. On the first message of a session, before anything else, the agent reads the three priorities. If no plan is set yet, the reminder says so and the agent offers to set one: it never invents one.
Proof: a check sets a fake plan in a test home and verifies that Claude Code, opencode and Antigravity all receive it on the first line — and that a day without a plan says so.
4. Why "Thought Communication"
Here, the tower is thought of as a role-playing game: an ability is unlocked by a test, and a skill's name says what it does. In Slime, Rimuru speaks to all of his people at once, by thought (思念伝達, Shinen Dentatsu): everyone knows what the leader has in mind, without a meeting. The skill has four moves:
- on waking, read the thought: the daily plan;
- sort every request: in the plan, "if time allows", or off-plan — and when it is off-plan, the agent says so and Patrick decides;
- never change the plan yourself: the agent reads, Patrick sets;
- report per priority, with proof.
Proof: the skill's own check verifies 12 points: every file it cites exists, the reminder really starts with the plan, the checks above are green, the wake-up skill points to it, and all three agents can see it.
5. Does anyone else do this?
Yes. Mission Control (open source, AGPL-3.0) is a task board for solo entrepreneurs who delegate to agents, with a daily plan. Soloop (a paid service) gives a solo founder a team of agent roles. solo-founder-os (open source, MIT) brings together ten agents, evaluation and a skill library. Our difference: every rule is held by code that refuses, and every refusal has its test.
The common rule: a plan nobody reads protects nothing. Here, every agent reads it on waking, and a test checks that this is true.