Adr Directory

Type: kb/types/index.md

← Parent

  • 002-inline-global-types-in-writing-guide (adr) - Decision to inline note and structured-claim templates into WRITING.md so the agent gets type structure and writing conventions in a single hop — eliminates one read for the two most common note types
  • 003-connect-skill-discovery-strategy (adr) - Design options and scaling strategy for how the connect skill discovers candidate connections — index-first with semantic search backup, and what changes when the KB grows
  • 004-Replace areas with tags (adr) - Replaces the areas frontmatter field with freeform tags and restructures index pages to have both curated and generated sections, decoupling navigation from comparative reading
  • 005-quality-check-placement (adr) - Where quality checks belong — WRITING.md, validate, the experimental review system, or recurring tasks — based on cost, false-positive tolerance, and whether the check gates note creation
  • 006-two-tree-installation-layout (adr) - Decision to split installed projects into two directory trees — user content in kb/, framework in commonplace/ — with operational artifacts copied to kb/ for fast agent access and methodology kept in commonplace/ as fallback
  • 007-reports-directory-for-generated-snapshots (adr) - Decision to create kb/reports/ for generated, regenerable analytical snapshots — distinct from workshop (temporal work-in-flight) and notes (durable claims)
  • 008-Stdlib-only core scripts (adr) - Core scripts use only Python stdlib by defining a strict frontmatter grammar that a regex parser handles completely
  • 009-Link relationship semantics (adr) - Adopts a fixed vocabulary of link relationship types (extends, grounds, contradicts, enables, exemplifies) borrowed from arscontexta and adapted for agent navigation under bounded context
  • 010-review state should move to sqlite once reviews leave git and accumulate operational metadata (adr) - Accepts SQLite as the canonical store for review state once review artifacts are removed from git and start carrying selector and ack metadata rather than just human-readable prose
  • 011-Notes must be accessible to external readers (adr) - GitHub Pages commits the KB to external readers who lack CLAUDE.md context — KB vocabulary needs inline glosses and definition links on first mention
  • 012-types-for-structure-traits-for-review (adr) - Types define structural requirements checked by validation; frontmatter type identifies artifact kind; traits route semantic review gates; type definitions split into prose template and machine-readable schema
  • 013-skills-first-delivery-with-core-local-type-split (adr) - Historical decision to move skills to top-level sources, keep KB Goals always-loaded, split types into core and local, and use dynamic type discovery for practitioner-defined types
  • 014-scripts-as-python-package-one-tree-model (adr) - Decision to ship operational scripts as an installable Python package (llm-commonplace), replace the two-tree layout with a one-tree init model, and use scaffold symlinks for zero-duplication seeding
  • 015-standardize-authored-type-definitions-on-json-schema (adr) - Accepted decision to replace the custom type-profile DSL with authored JSON Schema over a parsed note document model
  • 016-custom-types-use-template-instruction-pairs (adr) - Accepted decision that specialized and practitioner-defined types use separate template and instructions files; the former WRITING.md loading assumption is superseded by ADR-017
  • 017-collection-md-is-the-register-convention-boundary (adr) - Accepted decision to use collection-level COLLECTION.md files as the register convention boundary while keeping types structural and allowing mixed-register collections
  • 018-Types are path references to instruction docs (adr) - Replaces the enum type: field with a path reference to a single hand-authored instructions doc per type. Collapses the three-file-per-type layout (template / instructions / schema) into one authoring doc plus a sibling schema, and dissolves the filesystem-discovery, cross-collection ambiguity, and special-case routing problems that the current enum scheme forces onto the write skill.
  • 019-Collection-owned link vocabulary with per-destination outbound rules (adr) - Each COLLECTION.md owns its outbound-linking rules per destination collection, selecting from a shared authoring catalogue; the connect and write skills read COLLECTION.md directly, and the compiled register-by-register topology is retired
  • 020-Link vocabulary: add contrasts and mechanism; declare directional asymmetry (adr) - Adds contrasts and mechanism to the theoretical-register default template on corpus evidence, and declares the directional-asymmetry principle that governs how the theoretical vocabulary is authored and read.
  • 021-Ship library content under kb/commonplace (adr) - Ship Commonplace library content under kb/commonplace/ so user collections at kb/notes/, kb/reference/, kb/instructions/ stay user-owned while the shipped library sits alongside as a read-only dependency.
  • 022-active-vocabulary-and-write-path-first-mentions (adr) - Accepted decision to define active vocabulary in AGENTS.md while making first-mention glossing a cp-skill-write authoring behavior
  • 023-Quote-anchored citations for code-grounded reviews (adr) - Code-grounded reviews may anchor load-bearing claims to a verbatim source quote pinned to the reviewed commit, with no byte offsets or ids; resolution is a write-time check against the checkout while the standing validator checks only shape, because the reviewed source is not retained in the KB
  • 024-Schema severity is per-constraint, fail by default (adr) - Schema-violation severity is a property of the constraint (read off error.schema, keyed by an optional stable ruleId), and the schema fails by default — a constraint opts down to a warning explicitly; replaces the instance-path _FAIL_PATHS proxy that collided on shared paths and under-enforced required structure
  • 025-Complete generated indexes are build-time only; agents read curated heads plus scoped rg (adr) - Complete generated listings (the per-collection dir-index and the per-tag generated tail) retire from git and the agent read path and are regenerated at mkdocs build time for human readers, while agents discover via committed curated heads plus scoped rg; supersedes ADR 003's index-first connect discovery, and adds no new query command because scoped rg recovers what the retired index gave.
  • 026-Split index into a committed tag-readme type with enforced marks (adr) - Splits the index type into a committed tag-readme type (file -README.md) whose weight gates (8KB warn/16KB fail, bytes, global) are the type contract, with an enforced complete mark (membership check) and a designed covered_by coverage mark; enforcement is reactive via the validator and the knowledge splits meaning to AGENTS.md, mechanics to the validator, maintenance to the type spec.
  • 027-Package scaffold assets without source-tree symlinks (adr) - Replaces source-tree scaffold symlinks with explicit Hatch wheel force-includes plus source-checkout fallback resolution for Commonplace packaged scaffold assets
  • 028-Design proposals live in kb/reference/proposals; ADRs record implemented decisions only (adr) - Unadopted system designs live in kb/reference/proposals/ as notes with the design-proposal trait; the proposed ADR status is removed so ADRs record only implemented decisions; the YAGNI gap rule routes system gaps to proposals and transferable insights to notes
  • 029-Review execution unified on (note, gate) pairs (adr) - Review execution is unified on (note, gate) pairs: one sentinel grammar keyed by the full pair, one prompt renderer, one batch executor with a salvage failure policy; bundle and gate sweep become packing strategies over the same pair protocol
  • 030-Harness-facing seams: batch prepare/ingest endpoints and runner adapters (adr) - The review system's execution medium becomes pluggable at two seams: batch-granular prepare/ingest CLI endpoints let any external orchestrator (live agent, harness workflow) execute review batches, and runner adapters put each subprocess harness CLI behind one registry-backed interface