thrashr888/clue/clue-context

Prepare project context

Use clue context to retrieve an evidence bundle, then read the returned records before explaining the next step.

clue context "notebook sharing" --sources reminders,files --list Alchemy --directory /absolute/project/path --budget-bytes 12000

Use the project and scope the user supplied. If a repository path is unknown, find it or ask rather than guessing. Queries use lexical terms: choose distinctive task names or concepts and split unrelated topics into separate calls. There is no implicit Git, Cortex, or Alchemy database integration in this version.

The budget covers the UTF-8 JSON encoding of the returned results array, not tokens or the whole envelope. context_bytes is measured. Oversized records may be omitted; compare selected with available and retain links/IDs for a later focused read. Do not copy giant artifacts into agent context.

For meeting preparation, include --sources calendar,reminders,notes with explicit --days-back and --days-ahead bounds. Calendar retains structured start/end/all-day metadata; date words are not parsed into filters. Use the separate clue-sqlite workflow for a database the user selects.

Inspect source coverage and errors. Notes default to title/folder search; --note-bodies reads a bounded sample of note contents. Search ranking does not establish freshness, task priority, completion of an implementation, or that a release shipped. Verify such claims against the relevant source or repository.

Write a brief handoff based on returned evidence: objective, relevant decisions/tasks, known gaps, and the next concrete step. Keep that synthesis distinct from literal source excerpts. --ai --share-content is optional external reranking and requires authorization to send query/title/snippets; local context requires no API credential.

For local AI ranking use --provider ollama --model INSTALLED_MODEL, or --provider systemone --base-url http://127.0.0.1:8009 --model kev-latest. Add --ai on search/context/SQLite search; rank always runs inference. These loopback paths need no --share-content and no TypeSafe credential. Environment defaults are CLUE_PROVIDER, CLUE_MODEL, CLUE_BASE_URL; use CLUE_PROVIDER_API_KEY only for a self-hosted server's own key. Non-loopback servers and Ollama cloud models require --share-content. Local proxy servers must be configured to keep inference local. Never silently fall back to a hosted provider.

Ollama returns relevance.kind: generated_rating without probabilities/confidence. Native System One results use native_distribution; confidence is not a cross-provider accuracy guarantee. Model availability, JSON validity, and ranking quality are separate checks. Start comparisons with synthetic examples/candidates.json; inspect api or ai.metadata for the provider actually used. Allow a longer --timeout for cold local inference (up to 600 seconds).

Inspect clue config show for saved/effective provider defaults before assuming a backend. Global defaults are set with clue config set --provider NAME --model MODEL [--base-url URL]. Flags override environment, then saved defaults; a provider override drops another provider's saved model/endpoint. Config never authorizes remote sharing. Consult the repository evaluation report before treating a model as reliable: successful HTTP/JSON validation is not a ranking-quality test.