Skip to content

ollama_repo_brief

Repo brief. FLAGSHIP compound job. Produces an OPERATOR MAP of a repo: {repo_thesis, key_surfaces, architecture_shape, risk_areas, read_next, evidence, weak, coverage_notes, corpus_used}. Takes source_paths (typically README + key src entries + manifests + docs) and optionally a corpus for cross-cutting context. Corpus-sourced evidence items carry retrieval score. Optional corpus_min_evidence_score drops corpus chunks below the relevance floor before the model sees them. Not a research clone — research answers a specific question; repo_brief synthesizes orientation. Every key_surface and risk_area cites evidence. read_next is INVESTIGATIVE (files or sections to look at), never prescriptive fixes or refactors. Thin evidence → weak=true with coverage notes. If you pre-extracted claims via ollama_extract with a frame argument, drop off-topic items before assembling source_paths — there’s no in-brief topicality gate for path inputs.

Parameter Type Required Default Description
source_paths string[] yes Files the brief should read server-side. Typically README, key src/ entries, package/manifest files, docs. Claude does not preload them — that’s the context-saving shape.
corpus string no Optional: named corpus (e.g. ‘handbook’, ‘doctrine’) for cross-cutting context beyond the repo’s own files.
corpus_query string no Query used to pull chunks from the corpus. Defaults to ‘repo architecture and surfaces’ when unspecified.
per_file_max_chars integer no Chars per source file (default 20k).
max_key_surfaces integer no Cap on key_surfaces (default 8).
max_risk_areas integer no Cap on risk_areas (default 5).
max_read_next integer no Cap on read_next (default 8).
corpus_min_evidence_score number no Minimum retrieval score (0–1) for a corpus chunk to enter evidence. Hits below the floor are dropped before the model sees them, with a counted note in coverage_notes. Absent → no relevance filter.

The JSON Schema below is generated from the same zod schema the server validates against at the wire — it cannot drift from runtime behavior.

{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"source_paths": {
"minItems": 1,
"type": "array",
"items": {
"type": "string",
"minLength": 1
},
"description": "Files the brief should read server-side. Typically README, key src/ entries, package/manifest files, docs. Claude does not preload them — that's the context-saving shape."
},
"corpus": {
"description": "Optional: named corpus (e.g. 'handbook', 'doctrine') for cross-cutting context beyond the repo's own files.",
"type": "string",
"pattern": "^[a-zA-Z0-9_-]+$"
},
"corpus_query": {
"description": "Query used to pull chunks from the corpus. Defaults to 'repo architecture and surfaces' when unspecified.",
"type": "string",
"minLength": 1
},
"per_file_max_chars": {
"description": "Chars per source file (default 20k).",
"type": "integer",
"minimum": 1000,
"maximum": 200000
},
"max_key_surfaces": {
"description": "Cap on key_surfaces (default 8).",
"type": "integer",
"minimum": 1,
"maximum": 20
},
"max_risk_areas": {
"description": "Cap on risk_areas (default 5).",
"type": "integer",
"minimum": 1,
"maximum": 10
},
"max_read_next": {
"description": "Cap on read_next (default 8).",
"type": "integer",
"minimum": 1,
"maximum": 15
},
"corpus_min_evidence_score": {
"description": "Minimum retrieval score (0–1) for a corpus chunk to enter evidence. Hits below the floor are dropped before the model sees them, with a counted note in coverage_notes. Absent → no relevance filter.",
"type": "number",
"minimum": 0,
"maximum": 1
}
},
"required": [
"source_paths"
]
}

Every tool returns the standard envelope — tier_used, model, tokens_in/tokens_out, elapsed_ms, backend provenance, warnings. See Envelope & tiers.

Schema + handler: src/tools/repoBrief.ts · registration: src/index.ts


Back to the Tool Reference.