Repository navigation
Guide coding agents through sys1rust setup and use - #39
Merged
Merged
Conversation
krishhgg
force-pushed
the
install/2-agent-docs
branch
from
October 6, 2026 23:21
81f9238 to
32d6d46
Compare
krishhgg
force-pushed
the
install/2-agent-docs
branch
from
October 6, 2026 23:36
32d6d46 to
c5bd67d
Compare
AGENTS.md tells coding agents which of 3 skills to follow, the rules that hold for every task and where the code lives. The skills live in .agents/skills/, which Codex, Cursor and GitHub Copilot read, and .claude/skills/ links to them so Claude Code loads the same files. CLAUDE.md imports AGENTS.md and adds how to keep a server alive past a Claude Code session. sys1rust-setup checks the Mac, runs install.sh, downloads the model, starts the server on demand or as the LaunchAgent, sends the README's request and tells the user what runs and how to stop it. It also covers update, uninstall and the failures an install hits. sys1rust-use describes the /v1/systemone request, the 3 answer types, the 2 confidences, the errors, the models and the serve options, with curl and Python examples. sys1rust-develop covers the source build, the test suites in order and the repository's rules for answers, speed claims, formatting, docs and commits. Every flag, number and message comes from the code, the README or the install.sh in progress. The scratch clone, the Mac checks, the PATH grep and the CLI help were run as written. The README request, the curl one-liner, the Python ask() example and the spawn example ran against a real sys1rust serve, and the CPU test step passes. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The Install section now opens with a one-line prompt to paste into a coding agent, which then follows AGENTS.md. The manual route is the curl | sh one-liner, and the background service is install.sh --service. Update and uninstall take 1 line each. Every Homebrew mention is gone, including the tarball and quarantine steps, since install.sh downloads with curl and checks SHA256SUMS. The Releases page still has the tarballs. Try it now stops the service with launchctl bootout, the source build section points to the develop skill, and the repository layout lists install.sh, the agent files and packaging/. All relative links and anchors resolve. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Use a fresh scratch checkout and a curl fallback when Apple's Git tools are unavailable. Run the installed binary by its absolute path, keep the chosen port throughout setup, create the log directory, and bound manual startup. Cover native arm64 execution for the piped installer and explain its ownership registry and lock recovery. Guarantee subprocess cleanup in the client example and describe the request's null handling. Stop the ignored test suites on the first failure and distinguish bit identity from the existing 0.001 tolerance checks. All 21 shell and Python examples parse, local links and Claude skill symlinks resolve, and controlled Git and download failures leave the working directory unchanged. The installer and real service lifecycle checks passed. Co-Authored-By: Codex <noreply@openai.com>
Explain that the installer locks shared bin and LaunchAgent paths as well as the prefix. Tell setup agents to retry interrupted updates, and describe how service readiness requires the LaunchAgent PID to own the listening port. Checked the prose against install.sh, verified the Claude skill link resolves to the updated file, and ran git diff --check. The installer regression suite passes all 113 checks under sh and dash, and the real LaunchAgent interruption test recovers successfully. Co-Authored-By: Codex <noreply@openai.com>
Document that the request model convaiinnovations/laya uses the checkpoint already served, while english explicitly names the English checkpoint and the same repo id selects English on the CLI. Make manual setup and the Python-owned server example select loopback and typed-decisions explicitly. Remove inherited API keys and revisions only from the child environment so the default unauthenticated verification request and pinned model remain consistent. Reuse these commands for restarting and offline setup. Checked all 18 shell and Python examples and both JSON examples in the skills and AGENTS.md, local links, Claude symlinks and git diff --check. Ran the documented manual setup, verification request and Python subprocess example against the release bundle with conflicting host, model, API key and revision settings. All succeeded and stopped their servers. Seven live model-field cases matched the corrected rules. Co-Authored-By: Codex <noreply@openai.com>
krishhgg
force-pushed
the
install/2-agent-docs
branch
from
October 6, 2026 23:51
7cea6ad to
3b202d4
Compare
Explain that environment assignments belong to the interpreter after the pipe. Add explicit no-clone service commands for native and Rosetta shells, with LAYA_PORT immediately before sh or arch, and quote the chosen port in clone-based commands. Reproduced the old pipeline losing port 8001 to an inherited 8000. Downloaded a fixture script through curl and verified both documented pipelines under sh and dash retain port 8001, pass --service and run as arm64. All 20 shell and Python examples and both JSON examples parse; local links, Claude skill symlinks and git diff --check pass. Co-Authored-By: Codex <noreply@openai.com>
Read the saved LaunchAgent port into SERVICE_PORT so updating a manual install does not overwrite its chosen PORT. Explain which port to use when checking the service and retain the manual port for its restart and verification blocks. Executed the documented update under sh and dash with a missing LaunchAgent and manual port 8001, and with saved service port 8002. Both cases preserved the correct port. All 20 shell and Python examples and both JSON examples parse; local links, Claude skill symlinks and git diff --check pass. Co-Authored-By: Codex <noreply@openai.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Give users one line to paste into their coding agent:
Add
AGENTS.md, a Claude entry point and setup, use and develop skills in.agents/skills, with Claude skill symlinks pointing to the same files. The setup guide handles fresh Macs without Git tools, Rosetta, PATH shadowing, chosen ports in clone and piped installs and manual updates, manual or login startup, updates, uninstall and installer lock recovery. Manual setup selects loopback and typed-decisions explicitly and isolates inherited API keys and revisions to match its verification request. The API guide distinguishes request model aliases from CLI model selection, covers actual request validation and guarantees subprocess cleanup. The development guide stops tests at the first failure and distinguishes bit identity from the existing 0.001 tolerance checks.This PR builds on the installer PR and adds the agent prompt to its README. Merge the installer first.
Validation:
All 20 shell and Python examples and 2 JSON examples in the skills and AGENTS.md parse; relative links and Claude skill symlinks resolve.
Controlled Git and tar download paths enter the scratch checkout; failures leave the working directory unchanged.
Commands and request fields match the final installer, CLI and validation code. Test claims match
settings.rs.The installer passed 117 checks under both shells; its real service lifecycle, documented candidate check and 120-question smoke test passed.
The documented manual setup, verification request and Python subprocess example passed against the real bundle with conflicting host, model, API key and revision settings. All servers stopped cleanly. Seven live model-field cases matched the guide.
Downloaded fixture scripts through curl confirmed the native and arm64 no-clone service commands preserve port 8001 under sh and dash. The previous pipeline reproduced the port loss.
The update commands preserved manual port 8001 with no LaunchAgent and used recorded service port 8002 under both sh and dash.
This PR appears safe to merge; the update instructions now preserve the manually chosen port.
What we checked:
PORT. For a running service, both the guide and installer read its savedLAYA_PORT.Summary
Adds agent guides for installing, calling and developing sys1rust, with shared Claude skill links and a setup prompt in the README.
SERVICE_PORTfrom the manually chosenPORT.Reviews (4) · Last reviewed commit: "Keep the manually chosen port when updat..." · Reviewed by Greptile