Domain detector and first-impression reader. Invoked at the start of a /taste run when the input is ambiguous (user dropped mixed materials, or format/domain unclear). Outputs a single domain hint and a one-paragraph phenomenological first impression. Optional — skip when /taste is called with explicit domain context.
Panofsky pre-iconographic and iconographic layer analyst. Produces the first two strata of visual interpretation — concrete visible attributes (color, light, composition, material, lens) and cultural convention markers (period, setting, wardrobe, iconography). Does NOT perform iconological interpretation (that is…
Warburg Nachleben specialist. Reads taste-icon's output and traces the visual motif to the Warburg panel(s) it exemplifies. Proposes new panels when no existing panel fits. Explicitly forbidden from shallow style labeling or influence-graph speculation without visual evidence.
Lens-aware natural-language query over the vault. Injects the active lens's prompts.md + entitymodel into the LLM context so answers use lens-appropriate terminology (aesthetic-warburg → Pathosformel vocabulary; engineering-alexander → Alexander pattern vocabulary; general-zettelkasten → Bloom-indexed notes).…
Health audit across the vault. Reads schema.md as the contract and reports divergences — schema-version drift, missing auto-generated sections, stale AAT/Wikidata resolutions, expired prompt tests, unstable Warburg panels, glossary drift, broken links, orphan pages, and freshness violations. Produces an actionable…
Lens-aware cognitive reconstruction of foreign-language sources into the user's vault language. Not translation — rewrites content using the target language's conceptual vocabulary while preserving key original terms as bilingual anchors. The set of anchor types to preserve is per-lens (aesthetic-warburg keeps…
LENS EVOLUTION — behavioral matcher that proposes an academic theory (Warburg-grade rigor) whose structure matches the user's unselfconscious working pattern. If accepted, operationalizes the theory as a new candidate lens installed at /.agent/lenses/ /. Reads observer state built silently from prior /ingest, /ask…
Self-evolving vault driver. Reads wiki/insights.md (graph-computed topology + content gaps), surfaces top-priority gaps, and either suggests or autonomously executes the next lens-appropriate follow-up (/taste under aesthetic, /ingest under any lens, /check --migrate). This is Bab-ilu's answer to InfraNodus-style…
Lens-aware gap discovery for the vault. Runs community detection (louvain or leiden per lens) over the vault's bipartite graph, qualifies tier-1 cluster candidates per D2 spec thresholds, surfaces unbridged gaps, and writes candidate files to .agent/todos/. Delegates to tools/gaprunner.py which wraps the lens-aware…
Initialize a Bab-ilu v2.1 vault with a chosen lens. Interactive 5-question flow (or --lens= skip), creates wiki/ directory structure per the lens's entitymodel, copies the active lens config to .agent/lenses/active/, optionally imports the seed kit. Run once per fresh clone.
Lens-aware entry point for any raw material (image / video / URL / PDF / text). Auto-detects type, resolves the output tier0 directory from the active lens's entitymodel, and dispatches to the lens-appropriate downstream skill (aesthetic-warburg → /taste; engineering-alexander → pattern-extraction…
Structural + lens-aware validation for the vault. Runs v2.0 entity checks plus (with --lens ) authority-anchor coverage and tier-1-cluster consistency. --rebuild-graph delegates to graphanalyzer. Safe to run any time; non-destructive without --fix.
Reset wiki state to a clean scaffold by scope (wiki / raw / log / checkpoints / all). Useful during development or carefree restarts after a botched setup.
Lens-aware taste/analysis entry point. Under aesthetic-warburg, runs the Panofsky/Warburg pipeline on a visual input and produces an Aesthetic MD. Under engineering-alexander or general-zettelkasten, runs the lens's own taste contract — reports over the vault's tier-0/tier-1 distribution, not a creative deliverable.…
A project guide for AI coding agents, stored in an AGENTS.md file. It describes the Bab-ilu project's principles, required practices, and design direction.