cli / commands / knowledgehillclimb

hillclimb knowledge

Cross-search learning: cards distilled from finished searches

hillclimb knowledge COMMAND

hillclimb knowledge backfill

Distill cards from every finished search that lacks one.

Walks runs/ and bootstraps learning from pre-existing history.

hillclimb knowledge backfill

hillclimb knowledge consolidate

The sleep phase: generalize claims, rewrite playbooks.

Lifts multi-family claims up the concept hierarchy (mechanical) and rewrites per-concept playbooks (one agent call per qualifying concept, routing key consolidate). Playbook rewrites land as reviewable git diffs.

hillclimb knowledge consolidate [OPTIONS]

Options

optionwhat it does
--dry-runShow what would generalize / which playbooks would rewrite; no writes, no agent calls.

hillclimb knowledge distill

Run the LLM claims pass on a finished search.

Distills typed claims (entities, concepts) — or, with --backfill, extracts them across existing cards.

hillclimb knowledge distill [OPTIONS] [SEARCH]

Arguments

argumentwhat it is
SEARCHSearch ref (run-id/search-id), or latest optional

Options

optionwhat it does
--backfillExtract claims for every knowledge card that has none.

hillclimb knowledge graph

Explore the knowledge graph.

Default: the interactive TUI screen (zoom/pan/click, time scrubber); --stats prints a text summary.

hillclimb knowledge graph [OPTIONS]

Options

optionwhat it does
--statsPrint index stats instead of the TUI.

hillclimb knowledge live

Show the live cards concurrent searches in a run are sharing.

The discoveries a sibling's next operator would receive.

hillclimb knowledge live [RUN]

Arguments

argumentwhat it is
RUNRun id, or latest optional

hillclimb knowledge query

Read-only memory lookup (no model calls).

Also advertised to operator agents so they can consult accumulated knowledge mid-search.

hillclimb knowledge query [OPTIONS] TERMS

Arguments

argumentwhat it is
TERMSKeywords, e.g. 'gradient boosting' or a technique slug required

Options

optionwhat it does
--family TEXTRestrict claims to one problem family.
--limit INTEGERDefault: 5.
--jsonMachine-readable output.

hillclimb knowledge rebuild

Force-rebuild knowledge/graph.json from cards and registries.

The graph is a derived index — always safe to rebuild, never hand-edit.

hillclimb knowledge rebuild

hillclimb knowledge show

Render the prior experience a new search would receive.

Scoped to this target.

hillclimb knowledge show TARGET

Arguments

argumentwhat it is
TARGETProblem target, e.g. emflow://gefcom2014:solar required