@iowarp/clio-coder 0.3.0

package/src/domains/prompts/fragments/wiki/plan.md

@@ -0,0 +1,48 @@
+---
+You are planning a repository wiki. You are not writing it. Your entire job this pass is to
+improve one JSON file so that the writers who follow you each have a well-defined page.
+
+The file is at `{{planPath}}`. It already contains a complete, usable plan derived from the
+repository index. Read it, then write an improved version back to the same path. If you change
+nothing, the run proceeds with what is already there, so only make changes you can justify.
+
+What you may change:
+- `overview`: replace it with one paragraph saying what this repository is, what it does, and
+  what a newcomer needs to know first. This becomes the opening of the wiki. It is usually
+  empty when you receive it, and filling it is the single most valuable edit you can make.
+- Merge two pages whose subjects are really one subject.
+- Split a page whose subject is really several independent subjects, giving each a distinct
+  path and a distinct set of anchor sources.
+- Rename a page whose title or path does not describe what lives there.
+- Drop a page that documents nothing a reader would look for.
+- Add a page for a component, workflow, or contract the index did not surface as its own area.
+- Re-anchor a page's `sources` onto the files that actually define its subject.
+- Sharpen a page's `intent` so the writer knows what to cover and what to leave out.
+
+Rules:
+- Emit the complete JSON object every time, not a patch. Keep the exact shape you were given:
+  `{"overview": "...", "pages": [{"path": "...", "title": "...", "intent": "...", "sources": ["..."]}]}`.
+  Any other key is ignored. Progress tracking is added by the harness; do not invent status fields.
+- `path` is relative inside the wiki, ends in `.md`, and uses `/` for sections. Nesting is how
+  sections are made: `domains/dispatch.md` and `domains/agents.md` form a `domains` section.
+- Never name `quickstart.md` or any `index.md`. Those are generated after every run from the
+  pages you plan, so a page dedicated to navigation is wasted work.
+- `sources` are repository-relative paths to files that exist. They tell the page writer where
+  to start; they do not have to be exhaustive.
+- Do not target a page count. Depth follows what the repository contains. A substantial service,
+  package, domain, or cross-cutting workflow earns its own page; a thin one does not.
+- Judge the plan against one question: after all these pages are written, could an agent that
+  never reads this repository's source still answer how it works and change it safely? Where the
+  answer is no, that gap is the page you should be adding.
+
+Discipline:
+- You have a small budget and one deliverable. Spend most of it reading: the instruction files
+  named below, the manifests, the entry points in the digest, and enough of the largest areas to
+  judge whether the proposed grouping matches reality.
+- Use `code_nav` to navigate. Do not read files wholesale when an outline answers the question.
+- Never read `.env` files or other secret-bearing files, and never quote their contents.
+- Write the file once, near the end. Writing it repeatedly spends budget for no gain.
+
+Style for `intent` and `overview`: dense, factual, complete sentences, no marketing prose. Do not
+use the pattern "[noun] - [parenthetical clause]"; use a full sentence or a colon instead.
+---

package/src/domains/providers/models/cloud-models/alcf.yaml

@@ -0,0 +1,40 @@
+# ALCF (Argonne) inference-gateway model knowledge base.
+#
+# GPT-OSS models served through ALCF are already covered by the existing
+# openai-gpt-oss family in local-models/clio-local-coding-targets.yaml.
+
+- family: meta-llama-4-maverick
+  matchPatterns:
+    - meta-llama/llama-4-maverick
+    - llama-4-maverick
+  capabilities:
+    chat: true
+    tools: true
+    toolCallFormat: openai
+    reasoning: false
+    structuredOutputs: json-schema
+    vision: false
+    audio: false
+    embeddings: false
+    rerank: false
+    fim: false
+    contextWindow: 131072
+    maxTokens: 8192
+
+- family: meta-llama-4-scout
+  matchPatterns:
+    - meta-llama/llama-4-scout
+    - llama-4-scout
+  capabilities:
+    chat: true
+    tools: true
+    toolCallFormat: openai
+    reasoning: false
+    structuredOutputs: json-schema
+    vision: false
+    audio: false
+    embeddings: false
+    rerank: false
+    fim: false
+    contextWindow: 131072
+    maxTokens: 8192