docs: strip consensus to pure decisions, move research to red-team, add Ubuntu install prompt#88
Conversation
Reviewer's GuideRefactors the main architecture/consensus doc to focus on high-level decisions, pushes model/quant/benchmark research into the red-team document, expands implementation details for the daemon and safety layers, and adds a new Ubuntu 26.04 LTS install prompt for the Minisforum AI Pro-370 MiniPC. File-Level Changes
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
|
Warning Review limit reached
More reviews will be available in 50 minutes and 33 seconds. Learn how PR review limits work. Your organization has run out of usage credits. Purchase more in the billing tab. ⌛ How to resolve this issue?After more reviews become available, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available. Please see our Fair Usage Limits Policy for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (2)
📝 WalkthroughWalkthroughThe PR refreshes the agent-seed system documentation across three files. The main ChangesAgent Seed System Documentation
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Hey - I've found 1 issue, and left some high level feedback:
- In
ubuntu-install-prompt.md, avoid hard-coding the usernamenamikazand instead have the agent explicitly ask the user for their preferred username to keep the prompt reusable across setups. - The safety architecture described in
agent-seed-architecture.md(5 layers implemented, 19 blocked patterns, 4 protected paths) appears to conflict with some items still listed as gaps inagent-seed-red-team.md(narrow blocklist, missing disk monitoring, etc.); consider reconciling these so the current state vs. remaining gaps is unambiguous.
Prompt for AI Agents
Please address the comments from this code review:
## Overall Comments
- In `ubuntu-install-prompt.md`, avoid hard-coding the username `namikaz` and instead have the agent explicitly ask the user for their preferred username to keep the prompt reusable across setups.
- The safety architecture described in `agent-seed-architecture.md` (5 layers implemented, 19 blocked patterns, 4 protected paths) appears to conflict with some items still listed as gaps in `agent-seed-red-team.md` (narrow blocklist, missing disk monitoring, etc.); consider reconciling these so the current state vs. remaining gaps is unambiguous.
## Individual Comments
### Comment 1
<location path="ubuntu-install-prompt.md" line_range="13" />
<code_context>
+
+1. **Download Ubuntu 26.04 LTS ISO** --- from ubuntu.com/download/server. Tell me where to get it and how to verify the checksum.
+
+2. **Create bootable USB** --- on my current machine (tell me the command, I'll tell you what OS I'm on if needed). Use `dd` or `balena-etcher`.
+
+3. **BIOS setup** --- boot the MiniPC, enter BIOS (tell me the key), and set:
</code_context>
<issue_to_address>
**nitpick (typo):** Consider using the official "balenaEtcher" capitalization/name instead of "balena-etcher".
This matches the product’s branding and avoids suggesting there is a `balena-etcher` command-line tool.
```suggestion
2. **Create bootable USB** --- on my current machine (tell me the command, I'll tell you what OS I'm on if needed). Use `dd` or `balenaEtcher`.
```
</issue_to_address>Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.
|
|
||
| 1. **Download Ubuntu 26.04 LTS ISO** --- from ubuntu.com/download/server. Tell me where to get it and how to verify the checksum. | ||
|
|
||
| 2. **Create bootable USB** --- on my current machine (tell me the command, I'll tell you what OS I'm on if needed). Use `dd` or `balena-etcher`. |
There was a problem hiding this comment.
nitpick (typo): Consider using the official "balenaEtcher" capitalization/name instead of "balena-etcher".
This matches the product’s branding and avoids suggesting there is a balena-etcher command-line tool.
| 2. **Create bootable USB** --- on my current machine (tell me the command, I'll tell you what OS I'm on if needed). Use `dd` or `balena-etcher`. | |
| 2. **Create bootable USB** --- on my current machine (tell me the command, I'll tell you what OS I'm on if needed). Use `dd` or `balenaEtcher`. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the agent-seed architecture documentation, adds model and quantization research to the red-team document, and introduces an Ubuntu 26.04 LTS installation guide. Feedback on these changes highlights several inconsistencies: the LLM API endpoint port is incorrectly listed as 11434 instead of 8080 in multiple places; the GGUF provider table contains a logical contradiction regarding KLD rankings; the RAM usage estimates for the 32K context contradict other sections; and the installation guide contains missing sudo commands and a redundant Tailscale installation step.
| | Git | Time machine, revert mistakes | | ||
| | Filesystem + shell | Read, Write, Bash, Grep | | ||
| | Web search/fetch | Research capability | | ||
| | LLM API endpoint | localhost:11434/v1 | |
There was a problem hiding this comment.
The LLM API endpoint is listed as using port 11434. However, port 11434 is the default port for Ollama, whereas the architecture has explicitly switched to llama.cpp server (which defaults to port 8080). This should be updated to localhost:8080/v1 to match the runtime choice.
| | LLM API endpoint | localhost:11434/v1 | | |
| | LLM API endpoint | localhost:8080/v1 | |
| ────────── ──────── | ||
| SSH via Tailscale ──► Ubuntu 26.04 LTS | ||
| ├── Tailscale | ||
| ├── llama.cpp <- localhost:11434 |
There was a problem hiding this comment.
The ASCII diagram still references localhost:11434 for llama.cpp. This should be updated to localhost:8080 to align with the default port of llama.cpp server instead of Ollama.
| ├── llama.cpp <- localhost:11434 | |
| ├── llama.cpp <- localhost:8080 |
| | **Unsloth** (Dynamic 2.0) | 16.8 GB / 17.1 GB (UD-Q4_K_XL) | ~0.548 (best at Q4) | **1st** | Dynamic per-tensor allocation; best KLD at all bit levels; includes developer-role & tool-call fixes | Slightly larger at same quant label | | ||
| | **bartowski** (imatrix) | 17.98 GB | ~0.577 | **2nd** | Wide quant coverage; imatrix calibration | 1 GB bigger than Unsloth at same quant | | ||
| | **lmstudio-community** | ~17 GB | ~0.329 | **4th** | Official llama.cpp team | Never on Pareto frontier for Qwen 3.5 | | ||
| | **mradermacher** (i1 variants) | ~15.4 GB (i1-Q4_K_M) | ~0.260 | **3rd** | Fine-grained IQK quants | Requires ik_llama.cpp fork | |
There was a problem hiding this comment.
There is a logical contradiction in the GGUF Provider Comparison table. The table header states that for KLD 99.9%, 'lower=better'. However, Unsloth (~0.548) is ranked 1st and bartowski (~0.577) is ranked 2nd, while lmstudio-community (~0.329) and mradermacher (~0.260) are ranked 4th and 3rd respectively. If lower KLD is indeed better, the rankings or the KLD values themselves appear to be swapped.
| | UD-Q4_K_XL | 17.1 GB | ~19 GB | ✅ Lots of room | ~1.5% loss, SOTA Q4 | | ||
| | Q4_K_M | 16.8 GB | ~19 GB | ✅ | ~2-3% loss | | ||
| | Q5_K_M | 19.5 GB | ~22 GB | ✅ Comfortable | ~1% loss | | ||
| | **Q6_K** | **22.5 GB** | **~25 GB** | **✅ Fits** | **~0.4% loss (near-lossless)** | | ||
| | UD-Q6_K_XL | 26 GB | ~28.5 GB | ⚠️ Tight | ~0.2% loss | | ||
| | Q8_0 | 28.6 GB | ~31 GB | ❌ Too tight | Negligible --- doesn't fit | |
There was a problem hiding this comment.
The 'RAM Used (32K ctx)' column in the Quant Selection table lists values that only add ~2.5 GB of overhead to the model size (e.g., ~25 GB for Q6_K at 22.5 GB). This contradicts the 'Critical failures' section of the same document (lines 31-34), which states that the KV cache at 32K context alone takes ~6 GB and OS/services take ~3-4 GB (totaling ~9-10 GB of overhead). Please reconcile these values to ensure accurate hardware planning.
| - Run `update-grub` | ||
|
|
||
| 6. **Post-install** --- after reboot: | ||
| - `apt update && apt upgrade -y` |
There was a problem hiding this comment.
The apt commands in step 6 are missing sudo. Since the system is configured with a non-root user namikaz in step 4, running these commands without sudo will fail with permission errors.
| - `apt update && apt upgrade -y` | |
| - sudo apt update && sudo apt upgrade -y |
|
|
||
| 6. **Post-install** --- after reboot: | ||
| - `apt update && apt upgrade -y` | ||
| - Install: `build-essential`, `git`, `curl`, `wget`, `tailscale` |
There was a problem hiding this comment.
In step 6, tailscale is listed for installation via apt. However, the Tailscale repository is not yet configured at this stage, so this command will fail. Furthermore, step 7 explicitly installs Tailscale using its official bootstrap script, making this entry redundant. It should be removed from step 6.
| - Install: `build-essential`, `git`, `curl`, `wget`, `tailscale` | |
| - Install: build-essential, git, curl, wget |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 7
🧹 Nitpick comments (1)
agent-seed-red-team.md (1)
208-208: ⚡ Quick winSandbox guidance conflicts with earlier recommendation.
Line 208 says systemd hardening gives “zero-config sandboxing without Docker or bubblewrap,” but earlier guidance (Line 128 in this same doc section context) positions bubblewrap as the preferred sandbox. Clarify whether systemd hardening is baseline defense or a full replacement.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@agent-seed-red-team.md` at line 208, The sentence about "systemd ProtectSystem/ReadWritePaths" claiming “zero-config sandboxing without Docker or bubblewrap” conflicts with earlier guidance that recommends bubblewrap as the preferred sandbox; update the text to clarify that systemd hardening (ProtectSystem/ReadWritePaths, ProtectHome, etc.) is a lightweight baseline defense and not a full replacement for an untrusted-code sandbox like bubblewrap, and explicitly state that bubblewrap remains the recommended isolation tool for executing untrusted agents while systemd hardening should be applied as complementary, default protection.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@agent-seed-architecture.md`:
- Around line 93-103: In agent-seed-architecture.md the two unlabeled fenced
code blocks (the numbered steps block beginning "1. Read GOAL.md..." and the
ASCII diagram block beginning "Your laptop MiniPC") must be
annotated with a language to satisfy markdownlint MD040; change the opening
fences from ``` to ```text for both blocks (and similarly for the other
occurrence around lines 113-128) so the fences are labeled while preserving the
block contents and closing fences.
- Line 26: The doc references a "llama.cpp server" but uses the
Ollama-associated port localhost:11434; either explicitly document that 11434 is
a remapped/proxied port (add a note describing the proxy/remap and how to
configure it) or change all occurrences of localhost:11434 to the actual
llama.cpp API endpoint/port your setup uses; update every instance of
"llama.cpp" and "localhost:11434" in the document so the port and description
are consistent (e.g., adjust the entries around the current mentions of
localhost:11434 and the "llama.cpp server" text).
In `@agent-seed-red-team.md`:
- Around line 149-157: The KLD direction/ranking mismatch must be fixed: decide
whether KLD is "lower=better" or "higher=better" and update the table values and
the verdict accordingly; specifically, either invert the note "lower=better" or
correct the KLD numbers for the rows Unsloth (~0.548), lmstudio-community
(~0.329), and mradermacher (~0.260) so that the quality ranks (1st for Unsloth,
3rd for mradermacher, 4th for lmstudio-community) match the metric direction,
and make the concluding sentence about Unsloth Dynamic 2.0 consistent with the
corrected metric and ranks.
- Line 93: The doc references inconsistent chat templates: change the Line 93
usage of "qwen3.6-enhanced.jinja" to the canonical template used elsewhere (the
froggeric fixed template referenced via the --chat-template-file invocation) so
all docs use the same template string; update any mention of
qwen3.6-enhanced.jinja to the canonical template name and ensure
agent-seed-architecture.md and lines around 193 use that same template reference
so deployment uses a single canonical chat template.
In `@ubuntu-install-prompt.md`:
- Around line 29-31: Split the kernel-parameter step into two explicit phases:
(1) Temporary one-boot change — instruct the user to interrupt the boot/GRUB
menu and edit the boot entry (e.g., press "e") to append `amd_iommu=off
amdgpu.gttsize=40960 ttm.pages_limit=14680064` for this single boot; and (2)
Persistent change — after confirming the kernel parameters work, edit
`/etc/default/grub` and add the same flags to GRUB_CMDLINE_LINUX_DEFAULT, then
run `update-grub` to make them permanent. Mention both steps clearly and label
them "Temporary (single boot)" and "Persistent (post-install)" so users don’t
perform the wrong action at the wrong time.
- Around line 35-39: There are two conflicting Tailscale install methods: the
inline install list item "Install: `build-essential`, `git`, `curl`, `wget`,
`tailscale`" and the separate curl installer block "`curl -fsSL
https://tailscale.com/install.sh | sh`" / `sudo tailscale up`; pick and keep one
authoritative method (either remove `tailscale` from the install list and retain
the curl installer block, or remove the curl installer block and keep the apt
package in the install list), and update the remaining text so it references
only the chosen method and the `sudo tailscale up` step consistently.
- Line 39: Replace the unsafe "curl -fsSL https://tailscale.com/install.sh | sh"
instruction with explicit apt/keyring steps: instruct adding Tailscale's GPG key
and apt repository (instead of the pipe-to-sh), running apt-get update and then
apt-get install tailscale, and keep the post-install command "sudo tailscale
up"; update the line that currently contains "curl -fsSL
https://tailscale.com/install.sh | sh" so it describes the keyring + apt repo +
apt-get install workflow and then "sudo tailscale up".
---
Nitpick comments:
In `@agent-seed-red-team.md`:
- Line 208: The sentence about "systemd ProtectSystem/ReadWritePaths" claiming
“zero-config sandboxing without Docker or bubblewrap” conflicts with earlier
guidance that recommends bubblewrap as the preferred sandbox; update the text to
clarify that systemd hardening (ProtectSystem/ReadWritePaths, ProtectHome, etc.)
is a lightweight baseline defense and not a full replacement for an
untrusted-code sandbox like bubblewrap, and explicitly state that bubblewrap
remains the recommended isolation tool for executing untrusted agents while
systemd hardening should be applied as complementary, default protection.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro Plus
Run ID: ee4b9adb-e5f3-40f2-800c-a4a79ad59ea2
📒 Files selected for processing (3)
agent-seed-architecture.mdagent-seed-red-team.mdubuntu-install-prompt.md
| ## LLM Runtime | ||
|
|
||
| **llama.cpp server** on localhost (not Ollama). Ollama has documented memory leaks (1GB -> 64GB over 24h, issue #10132) and vendors a stale llama.cpp that's ~56% behind upstream on AMD. | ||
| **llama.cpp server** on localhost (not Ollama --- memory leaks, stale upstream). |
There was a problem hiding this comment.
Clarify the llama.cpp API port to avoid setup breakage.
This doc says llama.cpp (Line 26) but uses localhost:11434 (Lines 80, 118), which is commonly associated with Ollama. If 11434 is intentional, document the explicit remap/proxy; otherwise align to the actual llama.cpp endpoint throughout.
Also applies to: 80-80, 118-119
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@agent-seed-architecture.md` at line 26, The doc references a "llama.cpp
server" but uses the Ollama-associated port localhost:11434; either explicitly
document that 11434 is a remapped/proxied port (add a note describing the
proxy/remap and how to configure it) or change all occurrences of
localhost:11434 to the actual llama.cpp API endpoint/port your setup uses;
update every instance of "llama.cpp" and "localhost:11434" in the document so
the port and description are consistent (e.g., adjust the entries around the
current mentions of localhost:11434 and the "llama.cpp server" text).
| ``` | ||
| 1. Read GOAL.md + run scripts/improve + scripts/eval --json | ||
| 2. Inject SCAN marker (~300t self-check) | ||
| 3. Sliding-window summarization (>4KB -> collapse) | ||
| 4. Run DefaultAgent (max 25 steps, $0.50, 300s wall) | ||
| 5. git commit agent changes | ||
| 6. Update CHANGELOG.md, git commit again | ||
| 7. Schema-validate modified JSON files (auto-revert) | ||
| 8. Health check (3 checks) + auto-rollback on failure | ||
| 9. Adaptive sleep (30s-15min) | ||
| ``` |
There was a problem hiding this comment.
Add languages to fenced code blocks to satisfy markdownlint (MD040).
Both fences are unlabeled; please mark them (for example, text) so lint passes and rendering is clearer.
Proposed patch
-```
+```text
1. Read GOAL.md + run scripts/improve + scripts/eval --json
2. Inject SCAN marker (~300t self-check)
3. Sliding-window summarization (>4KB -> collapse)
4. Run DefaultAgent (max 25 steps, $0.50, 300s wall)
5. git commit agent changes
6. Update CHANGELOG.md, git commit again
7. Schema-validate modified JSON files (auto-revert)
8. Health check (3 checks) + auto-rollback on failure
9. Adaptive sleep (30s-15min)
-```
+```
-```
+```text
Your laptop MiniPC
────────── ────────
SSH via Tailscale ──► Ubuntu 26.04 LTS
├── Tailscale
├── llama.cpp <- localhost:11434
│ └── Qwen 3.6-27B Q6_K (CPU AVX-512)
└── agent-seed/
├── AGENTS.md, GOAL.md, daemon.py
├── scripts/ (eval, route, improve)
├── tests/smoke.sh
└── CHANGELOG.md
Observer: git pull -> read log Daemon: survey -> think -> do -> verify -> commit -> sleep
+ CHANGELOG -> detect stuck Restart=always + daily cron
-```
+```Also applies to: 113-128
🧰 Tools
🪛 markdownlint-cli2 (0.22.1)
[warning] 93-93: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@agent-seed-architecture.md` around lines 93 - 103, In
agent-seed-architecture.md the two unlabeled fenced code blocks (the numbered
steps block beginning "1. Read GOAL.md..." and the ASCII diagram block beginning
"Your laptop MiniPC") must be annotated with a language to
satisfy markdownlint MD040; change the opening fences from ``` to ```text for
both blocks (and similarly for the other occurrence around lines 113-128) so the
fences are labeled while preserving the block contents and closing fences.
| | Provider | Q4_K_M Size | KLD 99.9% (lower=better) | Quality Rank | Strengths | Weaknesses | | ||
| | --------------------------------- | ------------------------------ | ------------------------ | ------------ | ---------------------------------------------------------------------------------------------------- | ------------------------------------------ | | ||
| | **Unsloth** (Dynamic 2.0) | 16.8 GB / 17.1 GB (UD-Q4_K_XL) | ~0.548 (best at Q4) | **1st** | Dynamic per-tensor allocation; best KLD at all bit levels; includes developer-role & tool-call fixes | Slightly larger at same quant label | | ||
| | **bartowski** (imatrix) | 17.98 GB | ~0.577 | **2nd** | Wide quant coverage; imatrix calibration | 1 GB bigger than Unsloth at same quant | | ||
| | **lmstudio-community** | ~17 GB | ~0.329 | **4th** | Official llama.cpp team | Never on Pareto frontier for Qwen 3.5 | | ||
| | **mradermacher** (i1 variants) | ~15.4 GB (i1-Q4_K_M) | ~0.260 | **3rd** | Fine-grained IQK quants | Requires ik_llama.cpp fork | | ||
| | **steampunque** (mixed-precision) | N/A | N/A | N/A | Hybrid layer quants, all-K types (no IQ slowdown) | **No Qwen 3.6-27B release as of May 2026** | | ||
|
|
||
| **Verdict:** Unsloth Dynamic 2.0 leads the Pareto frontier at every bit level. lmstudio-community is measurably worse at the same quant label. |
There was a problem hiding this comment.
KLD metric and quality ranking are internally inconsistent.
Line 149 says “lower=better,” but Line 153 (~0.329) and Line 154 (~0.260) are lower than Line 151 (~0.548) while still ranked worse, and Line 157 concludes Unsloth leads at every bit level. Please reconcile either the metric definition, the values, or the ranking/verdict.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@agent-seed-red-team.md` around lines 149 - 157, The KLD direction/ranking
mismatch must be fixed: decide whether KLD is "lower=better" or "higher=better"
and update the table values and the verdict accordingly; specifically, either
invert the note "lower=better" or correct the KLD numbers for the rows Unsloth
(~0.548), lmstudio-community (~0.329), and mradermacher (~0.260) so that the
quality ranks (1st for Unsloth, 3rd for mradermacher, 4th for
lmstudio-community) match the metric direction, and make the concluding sentence
about Unsloth Dynamic 2.0 consistent with the corrected metric and ranks.
| 5. **Kernel parameters** --- before boot finishes, edit grub to add: `amd_iommu=off amdgpu.gttsize=40960 ttm.pages_limit=14680064` | ||
| - Edit `/etc/default/grub`, add to `GRUB_CMDLINE_LINUX_DEFAULT` | ||
| - Run `update-grub` |
There was a problem hiding this comment.
Clarify temporary vs persistent kernel-parameter flow.
Line 29 says to edit GRUB “before boot finishes,” but Lines 30–31 describe persistent post-install changes in /etc/default/grub. Please split this into explicit phases (temporary one-boot edit vs persistent config) to avoid incorrect execution order.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@ubuntu-install-prompt.md` around lines 29 - 31, Split the kernel-parameter
step into two explicit phases: (1) Temporary one-boot change — instruct the user
to interrupt the boot/GRUB menu and edit the boot entry (e.g., press "e") to
append `amd_iommu=off amdgpu.gttsize=40960 ttm.pages_limit=14680064` for this
single boot; and (2) Persistent change — after confirming the kernel parameters
work, edit `/etc/default/grub` and add the same flags to
GRUB_CMDLINE_LINUX_DEFAULT, then run `update-grub` to make them permanent.
Mention both steps clearly and label them "Temporary (single boot)" and
"Persistent (post-install)" so users don’t perform the wrong action at the wrong
time.
| - Install: `build-essential`, `git`, `curl`, `wget`, `tailscale` | ||
| - Enable SSH | ||
| - Note the IP address | ||
|
|
||
| 7. **Tailscale** --- `curl -fsSL https://tailscale.com/install.sh | sh` then `sudo tailscale up` (will give an auth URL to visit) |
There was a problem hiding this comment.
Remove conflicting Tailscale installation methods.
Line 35 installs tailscale via apt, while Line 39 installs again via curl ... | sh. Keep one authoritative method to prevent drift and troubleshooting ambiguity.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@ubuntu-install-prompt.md` around lines 35 - 39, There are two conflicting
Tailscale install methods: the inline install list item "Install:
`build-essential`, `git`, `curl`, `wget`, `tailscale`" and the separate curl
installer block "`curl -fsSL https://tailscale.com/install.sh | sh`" / `sudo
tailscale up`; pick and keep one authoritative method (either remove `tailscale`
from the install list and retain the curl installer block, or remove the curl
installer block and keep the apt package in the install list), and update the
remaining text so it references only the chosen method and the `sudo tailscale
up` step consistently.
| - Enable SSH | ||
| - Note the IP address | ||
|
|
||
| 7. **Tailscale** --- `curl -fsSL https://tailscale.com/install.sh | sh` then `sudo tailscale up` (will give an auth URL to visit) |
There was a problem hiding this comment.
🧩 Analysis chain
🌐 Web query:
What is Tailscale’s current recommended Ubuntu installation method for Ubuntu 26.04 LTS, and do they still recommend curl -fsSL https://tailscale.com/install.sh | sh for production/server provisioning?
💡 Result:
For Ubuntu 26.04 LTS (“Resolute Raccoon”), Tailscale’s current recommended installation method is to install the client from their APT repository (i.e., add Tailscale’s GPG key + apt repo, then apt-get install the package). The Tailscale “Install Tailscale on Linux” docs explicitly show that for distributions with apt (including Ubuntu-based distros), you can install using the automated install script, but they also state you can do manual installation via the “Tailscale Packages - stable track page” instead of using curl | sh for those who prefer not to. [1] Regarding whether they still recommend curl -fsSL https://tailscale.com/install.sh | sh for production/server provisioning: the docs still list that exact one-command install as the “apt-based” (mainstream) method on Linux, and they describe the script as the same script available on their Download page. [1] The same URL (tailscale.com/install.sh) is still in use and includes logic to add the APT repository and install the tailscale package. [2] So: yes, Tailscale still provides and documents curl -fsSL https://tailscale.com/install.sh | sh as an install method for Ubuntu servers (production included), but they do not force that approach; for production/server provisioning they acknowledge a manual APT-repo route via the Tailscale Packages stable track for users who prefer not to use curl | sh. [1] Concrete Ubuntu 26.04 APT-repo example from the stable packages page: - Add key: curl -fsSL https://pkgs.tailscale.com/stable/ubuntu/resolute.noarmor.gpg | sudo tee /usr/share/keyrings/tailscale-archive-keyring.gpg >/dev/null - Add repo: curl -fsSL https://pkgs.tailscale.com/stable/ubuntu/resolute.tailscale-keyring.list | sudo tee /etc/apt/sources.list.d/tailscale.list - Install: sudo apt-get update && sudo apt-get install tailscale [3]
Citations:
- 1: https://tailscale.com/docs/install/linux
- 2: https://tailscale.com/install.sh
- 3: https://dl.tailscale.com/stable/
Avoid curl | sh in baseline server provisioning; use Tailscale’s apt repo/keyring steps instead.
Tailscale still documents curl -fsSL https://tailscale.com/install.sh | sh as the Ubuntu (apt-based) install method, but they also provide a manual install path (add Tailscale GPG key + apt repo, then apt-get install tailscale). Update ubuntu-install-prompt.md line 39 to use the keyring/apt repo method instead of piping to sh.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@ubuntu-install-prompt.md` at line 39, Replace the unsafe "curl -fsSL
https://tailscale.com/install.sh | sh" instruction with explicit apt/keyring
steps: instruct adding Tailscale's GPG key and apt repository (instead of the
pipe-to-sh), running apt-get update and then apt-get install tailscale, and keep
the post-install command "sudo tailscale up"; update the line that currently
contains "curl -fsSL https://tailscale.com/install.sh | sh" so it describes the
keyring + apt repo + apt-get install workflow and then "sudo tailscale up".
Cleans up consensus doc to only contain decisions. Moves GGUF comparison, model benchmarks, and quant research to agent-seed-red-team.md. Adds ubuntu-install-prompt.md for the MiniPC setup.
Summary by Sourcery
Refine the agent-seed architecture doc to focus on decisions and runtime loop behavior, move detailed model and quantization research into the red-team document, and add a guided Ubuntu installation prompt for the Minisforum AI Pro-370 MiniPC.
Documentation:
Summary by CodeRabbit