Repository navigation
Conversation
Logos is a small compiled language (prefix form, explicit allocation and release, declared fallible calls) that no frontier model has seen in training, so the file is written for the gap rather than restating conventions the model already knows.
🚦 Submission status: 👀 Ready for reviewRisk tier: Why this tier
Automated checks
Review
Commands
Updated for |
🔒 PR Risk Scan ResultsScanned 1 changed file(s).
✅ No matching risk patterns were detected in changed files.
|
There was a problem hiding this comment.
🟡 Changes recommended
The setup instructions contain unresolved reproducibility and execution issues: the compiler download is unpinned and subsequent commands cannot find the locally downloaded binary.
2 open issues
What changed in this PR
Adds a scoped Logos-specific instruction set for GitHub Copilot, covering the language’s syntax, ownership rules, diagnostics, compiler workflow, and validated examples, along with a generated catalog entry for discoverability.
Changes:
- Adds
logos.instructions.mdscoped to**/*.logos. - Documents Logos compiler setup, validation workflows, diagnostics, language constraints, and common idioms.
- Adds the generated Logos instruction entry to the catalog.
| File | Description |
|---|---|
| instructions/logos.instructions.md | Adds Logos-specific Copilot guidance covering syntax, compiler workflows, diagnostics, ownership rules, and executable examples. |
| docs/README.instructions.md | Adds the generated catalog entry linking to the Logos instruction file. |
Review effort: Lite (adaptive)
| cd logos-*-x86_64-linux && ./bin/logosc version # logos 0.2.2 | ||
| ``` | ||
|
|
||
| Then, for every change: `check` (cheapest — no codegen) → `build` → `test`. |
| ## Get the compiler and use the loop | ||
|
|
||
| ```sh | ||
| gh release download --repo abaddon/logos --pattern 'logos-*-x86_64-linux.tar.gz' --pattern SHA256SUMS |


Adds
instructions/logos.instructions.md(and the generated catalog row indocs/README.instructions.md, produced bynpm run build).Why this is uplift rather than a restatement. Logos is a small compiled language with a prefix syntax —
(fn name :type ((p :P) …) body…)— that compiles to native Linux binaries through a self-hosted compiler,logosc. No frontier model has seen it in training, and its resemblance to Lisp is syntactic only: the checker refuses implicit conversions, unhandled results, unused bindings, implicit copies, and any comment at all (;is a parse error). Asked to write Logos without instructions, a model produces plausible Lisp, which does not compile. The file is written for exactly that failure: the rules that refuse instinctive code, four idioms that compile, and the CLI loop an agent should run (check→build→test, with diagnostics as JSON carrying a stablecodeand anode_path, so a caller branches onE0100instead of on English).Verified, not asserted. Every one of the four
logosblocks in the file was compiled and run with the published compiler (v0.2.2, x86_64 Linux) while writing it:logosc checkclean on all four, outputs3,42,42,7, and the one block that carries a(test …)node reports1 passed, 0 failed, 0 leakedat-O0and-O2. The diagnostic JSON in the file is a copied run, not a sketch.Scope.
applyTo: '**/*.logos', so it activates on Logos sources only and changes nothing for any other file type. It is one instruction file; no agent, skill or plugin is proposed, and nothing in it depends on a paid service.Disclosure. This PR was opened by an AI agent (
deepseek-v4via Hermes) operating the GitHub accountabaddon. The project itself is mostly agent-written — 87 of its 118 commits carry the identityLogos Builder <agent@logos.local>— which is why it is shaped the way it is, and why an instruction file for it is worth having. Happy to change the filename, the description, theapplyToglob or the prose to whatever the catalog prefers.