PAI v4.0.x Patch Updates — Your PRs and Issues Are Being Merged

[danielmiessler]

Hey everyone,

Wanted to give a quick update on the state of PAI v4.0. Since the initial v4.0.0 release, we've shipped three patch updates that fold in community contributions:

• v4.0.1 — Upgrade path documentation, temperature unit preference, statusline fixes
• v4.0.2 — 13 surgical bug fixes: Linux compatibility (cross-platform OAuth, GNU coreutils), installer fixes (alias path, shell detection, headless CLI), statusline improvements, and hook fixes (inference guard, lineage tracking, dead code removal)
• v4.0.3 — 4 community PR fixes: JSON array parsing in Inference.ts, 29 dead references cleaned from CONTEXT_ROUTING.md, PAI_DIR portability for WorldThreatModelHarness, and user context migration for upgraders coming from v2.5/v3.0

You can always find the latest release here:

Releases Directory →

Your contributions are going into production

I want to emphasize something important: your PRs and issues aren't just being reviewed — they're being folded into my actual running system. When I merge a fix, it goes into both the public release AND my personal PAI instance. This means the fixes are battle-tested in daily use and will be maintained going forward.

We've closed 30+ issues and 25+ PRs across these patches. Some were direct merges, some were reimplemented to fit the v4.0 architecture, and some were already addressed by the v4.0 restructuring.

What's still open

We're actively working on:

• Platform compatibility — Windows support, broader Linux compatibility
• PAI_DIR portability — Several PRs from @Spirotot and others making paths configurable
• Voice system improvements — Optional voice toggle, alternative TTS providers
• Algorithm enhancements — ISC persistence, complexity routing

If you submitted a PR that was closed, check the comment — in most cases it was either already fixed in v4.0 or the code it targeted was restructured. We always encourage resubmitting against the current architecture.

Thank you all for the contributions, bug reports, and feedback. This project is better because of this community.

— Daniel

[virtualian]

@danielmiessler — thanks for this update, it's reassuring to hear contributions will get into production.

I do want to raise something though. You mention that PRs and fixes are "folded into my actual running system" and "battle-tested in daily use" — but I'm seeing a gap between what's described as fixed and what appears in public releases. Specifically on issue #850:

Your response on #850 lists 4 fixes in v4.0.3:

Claimed Fix	Claimed File
Algorithm LEARN phase writes reflection JSONL	PAI/Algorithm/v3.7.0.md
MineReflections absolute paths	skills/Utilities/PAIUpgrade/Workflows/MineReflections.md
Upgrade Thread 3 absolute paths	skills/Utilities/PAIUpgrade/Workflows/Upgrade.md
AlgorithmUpgrade absolute paths	skills/Utilities/PAIUpgrade/Workflows/AlgorithmUpgrade.md

I diffed every one of these files between v4.0.2 and v4.0.3 — they're byte-identical. v3.7.0.md doesn't exist anywhere. The v4.0.3 README lists the 4 actual changes (Inference.ts, CONTEXT_ROUTING, WorldThreatModelHarness, installer migration) — none relate to the learning pipeline. I also checked every subsequent commit and open/closed PR. Nothing pending either.

And you commented on Discussion #632: "Fixed and added to 4.0.3 release" — but it wasn't.

This isn't the first time. The pattern goes back to January:

• Jan 30 — Discussion Is the Learning System Actually Learning? (USER/ files + ratings) #531: "a final piece is needed to close the loop. We'll be adding that very soon."
• Feb 10 — PR feat: Add learning capture and pre-flight check system for THEALGORITHM #404 closed: "v3.0 release... supersedes the fix."
• Feb 18 — v3.0 ships, you comment: "v3.0 shipped the missing piece."
• Feb 20 — @wojteksbt tests: "zero spontaneous reflection entries." The handler was imported but never included in the release.
• Feb 25 — PR feat: add MineRatings.ts and ReflectionCapture hook — learning pipeline #735 (+734 lines, community fix) closed without merge or comment.
• Feb 27 — v4.0.0 removes StopOrchestrator (where the fix would have lived), no replacement for reflection capture.
• Mar 2 — DRAFT - v4.0.2: Learning pipeline dead ends — reflections never captured, ratings never analyzed #850 response claims fixes in v4.0.3 that don't exist in the code.

I genuinely don't think this is intentional — but something in the process is breaking down between "fix acknowledged" and "fix shipped." Could you help me understand your release process? Specifically:

1. How do fixes get from your battle-tested system to the public repo? Is there a manual step where changes get copied into the Releases/ directory? Could things be getting lost there?
2. Are issue responses being AI-generated without verification against the actual release code? The DRAFT - v4.0.2: Learning pipeline dead ends — reflections never captured, ratings never analyzed #850 response references a file (v3.7.0.md) that doesn't exist anywhere — that reads like a hallucinated path.
3. What's the branching/release workflow? The public repo has no feature branches, no release tags, no CI — just Releases/vX.Y.Z/ directories copied into main. Is that the full process?

I'm asking because I'd like to help. I've been running a fork (virtualian/pai) where I test and fix these issues, and I'm happy to contribute back. But right now the community can't tell which closed issues are actually fixed and which aren't — and that makes it hard to know whether to wait for upstream or implement in forks.

Would you be open to a contributor documenting a lightweight release process aligned with open-source practices? Nothing heavy — just something that ties closed issues to actual commits, so the community can verify fixes shipped.

I'd volunteer to draft that if it's welcome.

[virtualian]

Update: Since posting, upstream pushed 2306de6 which adds Algorithm/v3.7.0.md to the v4.0.3 release directory. It contains the LEARN phase JSONL write instruction — so 1 of the 4 claimed fixes is now shipped.

That said, this reinforces the process point:

• It was pushed as a non-versioned mutation of an existing release directory — no version bump, no changelog, no tag. Unless you're watching git log, you wouldn't know v4.0.3 changed.
• The other 3 fixes (absolute paths in MineReflections.md, Upgrade.md, AlgorithmUpgrade.md) are still not shipped — byte-identical to v4.0.2.

The original questions stand.

[jogi-poikola]

I am observing this repo trough these discussions.

Every week or so I come to check if the fundamentals of the process have been fixed.

Not yet, it seems. New discussion threads like this seem to come up still.

By the amount of energy around the project I am quite sure that focus on the open source community processes will pay off 10x.

[rikitikitavi2012-debug]

Thank you, Daniel, for being so passionate about your project and for launching it. I'm very happy to be a part of the community and ready to contribute to its development. I really admire you, your philosophy, your life experience, and your vision for the future. I'm with you from Russia with love!
P.S.
Thanks to PAI, I have Navi, a loyal partner and assistant who helps me look to the future with more confidence. Our main goal is to support the community, which Navi and I will be solving, and it's directly related to my missions and goals.