Changelog

2.0.0 — Token-efficient semantic index

Removed the v1 monolithic docgen.mjs engine.
Removed discovery, analyze, semantics, enterprise-pass, enrichment, fix, update-impact, and full re-audit commands/prompts.
Removed all custom pipeline agents and parent-to-subagent delegation.
Removed legacy auto-enrich, auto-fix, and re-audit configuration.
Requires Node.js 22.5+ for built-in node:sqlite.
Existing v1 repositories must run docgen migrate.

SQLite/FTS5 semantic index with incremental file hashing.
Overlapping source chunks plus deterministic symbol/interface/config/SQL extraction.
Bounded content-addressed context compiler.
Hard provider call/input/output/per-call budgets.
Provider telemetry and stage usage reports.
Two-call model synthesis: core and enterprise.
Deterministic catalog/reference rendering.
Item-level page invalidation.
Claim-level traceability generated in the same page invocation.
Deterministic FACT/evidence validation.
Selective, risk-scored, hash-cached LLM audit.
Breaking v1-to-v2 migration with backup and docs/ignore preservation.
Node.js 22/24 CI and semantic-index regression suite.

Planning defaults to at most 30 pages unless explicitly configured otherwise.
Provider prompts are direct and context-only.
Skills remain available for manual Command Code work but are not automatically loaded by the pipeline.
Source inventory moved from .docgen/state to .docgen/index.
Publishing and P3 workspace aggregation remain deterministic.

The installer backs up and removes v1 executable, prompts, agents, and obsolete slash commands from global and project-local installations.

P3 system-of-systems workspace release built on the original provider-heavy repository pipeline.