Use for a quick, single-paper formalization into a Gaia knowledge package: read one academic paper (Markdown preferred; plain-text or other readable formats also accepted) and emit a standalone `<name>-gaia/` package. Runs a four-phase analytical workflow (Phase 1 extract conclusions / motivation / open questions / cross-conclusion logic graph; Phase 2 reconstruct each conclusion's reasoning chain; Phase 3 audit weak points and highlights, calibrate leaf priors; Phase 4 emit Gaia DSL package files), gated by an upfront suitability check (skip review/survey/perspective papers and corrupted paper text). Surfaces 9 argument-pattern weak-point types (`measurement`, `causal`, `model`, `statistical`, `generalization`, `comparative`, `formal`, `computational`, `external`). Cross-grounds the paper against LKM's existing knowledge graph in Phase 1b via `gaia search lkm knowledge`, filtering on `provenance.source_packages` and verifying reasoning-chain closure via `gaia search lkm reasoning --claim-id` (graph-shaped by
Use when formalising a knowledge source (scientific paper, textbook chapter, technical report) into a Gaia knowledge package and you want the thorough, audit-grade treatment. Walks a six-pass pipeline — extract → connect → check completeness → refine strategy types → verify structural integrity → polish for standalone readability — with a compile + check loop after every pass and a prior-assignment + inference tail. Emits the package source, `priors.py`, and `ANALYSIS.md`. This is the **Paper → package** entry point when depth matters: the slow, exhaustive sibling of `gaia-formalize-coarse` (which is a quick four-phase single-pass for one paper). Reach for `gaia-formalize-fine` when the source is load-bearing, multi-section, or destined for publication; reach for `gaia-formalize-coarse` for a fast first cut of a single paper.
Use when assigning priors to independent claims in a Gaia knowledge package via `priors.py` and `register_prior`. Carries the prior-assignment guide (evidence-level → prior-range table, abduction π(Alt) explanatory-power semantics encoded as the `p_e_given_h` / `p_e_given_not_h` likelihood ratio) and the iteration loop. Different from `gaia inquiry review` — that is a graph-health publish-gate verb; this skill is the prior-assignment workflow.
Use when the user wants a browsable Obsidian vault (`gaia-wiki/`) from a Gaia knowledge package. Generates a skeleton via `gaia run render --target obsidian`, then rewrites every page to faithful-reproduction depth (full derivations, figure embeds, Review/Supports/Significance/Caveats sections), authors narrative section chapters in DSL module order, and specialises Weak Points + Open Questions sections.
Use after `gaia run render <pkg> --target github` to fill the README skeleton with a per-conclusion evidence-assessment narrative, write Weak Points and Evidence Gaps sections framed around internal nodes, then push README + ANALYSIS.md + docs/detailed-reasoning.md to GitHub. Agent-side prose discipline anchored on a single render verb.