chirality-app

Read one deliverable folder and produce a structured 7-section content digest summarizing identity, scope, document kit, dependencies, references, semantic framework, and quality observations.

Transform raw concordance evidence atoms and risk inventory rows into typed publication concordance candidates for register freeze review.

Post-synthesis cross-section semantic concordance verification for rewritten DBM package outputs.

Review a human-prepared draft DBM against the governed knowledge base — build review substrate, prepare candidate findings for human disposition.

Post-authoring evidence bundle and review for pipeline-produced DBM sections — build review substrate, prepare candidate findings for human readiness judgment.

Assemble the full rewritten DBM package from approved section outputs and produce package QA artifacts.

Publish exactly one rewritten DBM section from approved mapped DOMAIN inputs and emit fixed QA artifacts.

Review one DOMAIN decomposition package as a closed system for derivative parity, active snapshot completeness, and handoff-state readiness. Use when a root-level remediation or scope-change run needs package-level closure evidence before downstream work or phase advancement.

Run a deliverable-local consistency sweep across production documents in one deliverable. Use when checking for unresolved markers, missing core artifacts, identity-label mismatches, or candidate unsourced numeric statements.

Extract dependency register (Dependencies.csv v3.1) from deliverable source documents using Anchor x Execution edge typing with evidence-first provenance. Setup-pipeline skill dispatched by ORCHESTRATOR during project setup.

Draft knowledge-artifact document set for a DOMAIN Knowledge Type deliverable. Produces KA-* files per the variable artifact schema. DOMAIN pipeline only.

Independently re-extract one rasterized PDF page's prose, equations, and asset placeholders for downstream extraction-reproducibility comparison (Gate 1.5-P Stage 1).

Per-dispatch-unit bounded atomization of one slice of one DOMAIN source markdown into IN/OUT/TBD atomic units with dual SourceRefs, content hashes, and vocabulary seeds. Invoked by the DOMAIN_DECOMP orchestrator for per-skeleton-dispatch-unit fanout in Phase 2.

Per-page bounded extraction of structured data from a single engineering drawing page image, using crop-first multiblock VLM reasoning. Drawing-type-aware (PFD implemented; P_AND_ID, ISOMETRIC, GA stubbed). Invoked by the DRAWING_EXTRACT orchestrator for per-page fanout.

Per-page titleblock extraction from drawing sheets using four corner crops and a full-page thumbnail; emits a governed sheet-inventory stub.

Detect display-equation bounding boxes on one rasterized page PNG using multimodal vision and emit normalized coordinates for downstream cropping. Per-page bounded dispatch invoked by EQUATION_AUDIT during Phase 1 when per-equation crops are enabled.

Convert one human-written natural-language correction note about one display equation into a corrected LaTeX expression. Per-flag bounded dispatch invoked by EQUATION_AUDIT during Phase 3a, one TASK per flagged entry whose `description` is prose rather than LaTeX.

Extract costing-relevant design basis parameters for equipment matching a target module type list from KA files in a single KTY folder. Use when building equipment costing registers for cost estimation.

Extract discrete physical equipment items from Knowledge Artifact files within a single DOMAIN Knowledge Type folder. Use when building per-KTY equipment registers from KA production documents.

Generates estimation input package (pricing library, LOE, project parameters, BOE) via two-phase run with PHASE=SCAFFOLD|BOE modes and a human gate between invocations.

Generate an immutable, provenance-first estimate snapshot for a bounded scope of deliverables and/or packages, writing quarantined artifacts under the estimating tool root. Use for automated estimate runs driven by a validated BASIS_OF_ESTIMATE enum and local pricing sources.

Draft and iteratively enrich the four-document kit (Datasheet, Specification, Guidance, Procedure) for a PROJECT or SOFTWARE deliverable. Dispatched with RUN_PASSES parameter.

Archive, tombstone, verify, and emit disposition evidence for DOMAIN KTY content after an approved scope change without rewriting active Knowledge Artifact content.

Review and optionally align one DOMAIN Knowledge Type folder's metadata surfaces to approved decomposition truth without rewriting Scoping.md or KA content. Use after regeneration or scope change when `_CONTEXT.md`, `_STATUS.md`, or `_REFERENCES.md` may have drifted.

Generate a matrix-organized, coverage-complete semantic lensing register (_SEMANTIC_LENSING.md) from _SEMANTIC.md + production documents. Setup-pipeline companion to the interactive semantic-lensing skill.

Extract pixel-anchored P&ID valve symbol instances from prepared tiles using isolated-crop classification before any row may count.

OBSOLETE legacy per-tile P&ID valve candidate extraction retained for historical readability; new runs use pandid-valve-symbol-instance.

Read one rasterized PDF page and emit the visible printed folio label (page number as it appears on the page itself — "47", "xiv", "B-3", or null when no folio is printed).

Identify visible figures, tables, and meaningful images on one rasterized prose PDF page and emit bbox-normalized asset records for deterministic materialization.

Single-page merged VLM pass that produces BOTH the raw Markdown transcription and the bbox-normalized asset JSON from one vision read of the page image. Replaces the prior two-skill split (pdf2md-page + pdf2md-page-assets) for new dispatches.

Convert a single PDF page PNG to raw Markdown via multimodal vision. Use when PDF2MD orchestrator spawns per-page fanout; one image in, one markdown file out, no post-processing.

Convert a bounded PDF source into Markdown by combining deterministic PDF tooling (rasterize, post-process, assemble) with the agent's multimodal vision for per-page transcription. Use for single-run TASK-shell conversions where the full PDF2MD persona orchestration is unnecessary.

Structured recommendation output using PROPOSAL blocks with evidence, change, risk, and status fields. Use for any deliverable-local task that surfaces actionable findings.

Apply semantic lensing analysis to deliverable documents using _SEMANTIC.md matrices. Treats lensing entries as a candidate worklist, tags proposals with matrix coordinates, and optionally updates lens artifacts.

Generate the deliverable-local semantic lens (_SEMANTIC.md) by adopting canonical matrices A and B and deriving matrices C, F, D, K, G, X, T, E via semantic algebra, showing all interpretation work.