Skip to content

jscraik/Agent-Skills / Plugins/aidevcon/skills/talk-maleix-collective-intelligence

talk-maleix-collective-intelligence

Verified

Provides detailed answers, conceptual explanations, workflow guidance, and framework-based analysis about Edouard Maleix's talk "How AI-First Dev Teams Build Collective Intelligence — One Attributed Mistake at a Time." Use when the user asks about giving coding agents their own identity and signed commits, the diary/entry/pack/render workflow, turning agent mistakes into reusable team knowledge, evaluating knowledge packs for fidelity and usefulness, voluntary task picking by autonomous agents, the MoltNet open-source project, compound engineering, or applying his approach to make agent lessons compound across a team instead of evaporating into chat history.

FileKindSize
SKILL.mdskill md9.5 KB
agents/openai.yamlconfig140 B
outline.mdmarkdown10.8 KB
quote.mdmarkdown4.3 KB
transcript.mdmarkdown27.8 KB
Topic-tagged verbatim quotes (with approximate transcript section). Speech-to-text artifacts preserved as-spoken; bracketed gloss added when meaning is unambiguous.

### On the core problem (lessons evaporating)
1. *"All the small discoveries, all the tiny learnings, all the incidents that are inside the session. They rarely make it out of it. They just end up closed here. So correction evaporates."* — The familiar obstacles
2. *"Agent work create lessons every day. Like you have those leisurely in the session, but the team is not learning."* — What we actually need
3. *"What our agents learned yesterday that our team can use today or tomorrow and the days after."* — Opening
4. *"What is your agents learned yesterday that your team still knows today?"* — Closing (the inverted closing form of the opening question)

### On rules / skills accumulation
5. *"We have lots of rule that buys up. Some skills, some project notes. And yeah, the dress pile up, same as your context below. It really gets crowded. And you barely can barely say when it's getting used, if it's still useful, if it became stale, and even if it gets activated when you expect it to be activated."* — The familiar obstacles
6. *"The PR can show up green, like all test passes, … you can hardly, you cannot even tell actually what lessons shape that work."* — The familiar obstacles

### On accepting knowledge decay
7. *"You will let some of those guidance fail because some knowledge just decay. We have to accept that."* — What we actually need
8. *"Instead of having. Static documentation, we have something that constantly move die the day after. Not a problem. Just need to accept that."* — What we actually need

### On identity (Act 1)
9. *"PR was really written, I think 95% by globally [Claude]. … And all the commits are under my name. They are even my GPT [GPG] signature."* — The Testify PR anecdote
10. *"Give the agent an identity. That's the first primitive. … The identity gives us tier separate actor. It gives us some sign commits."* — Identity primitive
11. *"It's even worse for the PR reviews. … you might have received some PR reviews. Authored by one of your teammates, but actually counter by global [Claude]."* — The Testify PR anecdote

### On the diary
12. *"The place where the stops being just the work and the lessons start being a four-way [forward] artifact."* — The Diary primitive
13. *"If the work or decision matters, you need to lend it somewhere before it disappear."* — The Diary primitive

### On rendering with attribution
14. *"Every section can see points through given entry. … Keep the attribution live. So here you can see the source with the answer identifier. The human that was operating. And the agent identifier as well."* — Render
15. *"Not have just the skill that says that's how you do things. It's more like that's how you do things because we got beaten somewhere that hurts and we don't want that to happen again."* — Render

### On the pipeline (Act 2 summary)
16. *"Interruption becomes an entry. The right entries become a pack. Then the pack becomes something the agent can read. And then in the end that one developer paid for … We counted team asset."* — Act 2 summary
17. *"There is even a discipline. I did not hear the word yet today. It seemed to be called compound generating [engineering]."* — Act 2 summary

### On evals
18. *"Before you run an llm judge you are the judge. You do the work that the judge will do yourself."* — Evals: Fidelity
19. *"For that go plan scenario without the pack we fail at 67 every time the go sdk was missing from the agent work. Whereas with the back it was always passing even after repeating the same task."* — Evals: Usefulness

### On autonomy
20. *"You as the human keep the goal, the judgment and the responsibility and the agents keep the continuity repetition."* — Act 3
21. *"You just have agent with some set of capabilities. And any of those agents can fire a task if it matches its capability. So it pre-picks voluntarily."* — Act 3 (voluntary task picking)

### On curation (Q&A)
22. *"Every workflow before we should lend it to an llm. We need to master it as human because then you need to explain that llm how we issue two things and if you are not able to explain it how can we expect the length [LLM] to be for you."* — Q&A on curation
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 323kUpdated Claude CodeCodexVerifiedInstall