Changelog

Subscribe via RSS

v2.6.0 — February 15, 2026

Templates are now schema definitions — the single source of truth for task, project, and all vault primitives.

Template Schemas

Templates define YAML schemas with typed fields, defaults, enums, and required markers. createTask() and createProject() read from templates at runtime — no more hardcoded field lists.

# templates/task.md frontmatter
schema:
  status:
    type: string
    required: true
    default: open
    enum: [open, in-progress, blocked, done]
  priority:
    type: string
    required: true
    default: medium
    enum: [critical, high, medium, low]
  due:
    type: date
    description: Target completion date
  tags:
    type: list
    description: Categorization tags
  estimate:
    type: string
    description: Time estimate (e.g. 2h, 1d)

Advisory Validation

New validateFrontmatter() checks required fields and enum values against schemas — warns but never blocks. Agents create tasks fast; validation catches drift.

Vault Templates Override Builtins

Drop a template in your vault's templates/ directory and it takes precedence over builtins. Full malleability — customize any primitive's schema without forking ClawVault.

Improved

buildTemplateIndex() scans vault templates first, falls back to builtins
16 builtin templates ship with schema fields
clawvault init generates .base files compatible with new schemas
459 tests passing, typecheck clean

v2.5.4 — February 15, 2026

Fixed

Hardened cross-platform qmd result path normalization so search and context scoring behave consistently on Windows and Unix.
Fixed WebDAV path safety/root containment handling to remove false 403s and improve secure path resolution on Windows.
Stabilized cross-platform test setup for WebDAV temp vaults and shell-init path expectations.

Improved

Reliability baseline now validates cleanly with full passing suite after path portability fixes (449/449).
Updated skill guidance to emphasize runtime verification and integration checks (openclaw --version, clawvault compat).

v2.5.3 — February 15, 2026

Fixed

Consolidated setup guidance into one canonical OpenClaw runtime path (hooks install + hooks enable + explicit verification).
Removed CLI/docs drift across installation, compat, context, wake/sleep, graph, qmd integration, and OpenClaw lifecycle docs.
Clarified skill-vs-hook distinction and added direct troubleshooting for hook discovery precedence conflicts.

Changed

Updated command references to match current runtime flags/options (compat --json, graph --json, context --max-hops, sleep --session).
Simplified docs content to executable workflows and removed speculative/undocumented flags and config keys.
Added safer AGENTS.md guidance: append ClawVault instructions instead of replacing existing prompt/bootstrap behavior.

v2.5.0 — February 15, 2026

Two headline features that change how agents interact with their vaults.

Dynamic Prompt Injection

New clawvault inject command pulls relevant decisions, preferences, and rules directly into prompt context. Two-layer matching:

Deterministic matching (default) — keyword and scope-based rules, zero latency
LLM fuzzy matching (opt-in --enable-llm) — classifies message intent, finds relevant vault entries

clawvault inject "How should we handle the Hale deployment?"
clawvault inject --enable-llm "What's our pricing strategy?"

First-Class Project Primitive

clawvault project with full lifecycle: add, update, archive, list, show, tasks, board.

clawvault project add "Site Machine" --owner pedro --client "Hale Pet Door" --status active
clawvault project tasks site-machine
clawvault project board --group-by client

Projects link to tasks, generate Obsidian Kanban boards, and support filtering by status/owner/client.

Also Added

Pluggable compression backends — Ollama, Minimax, GLM, any OpenAI-compatible backend.
Centralized transition logging — task state changes write to transitions.jsonl ledger.

Improved

Canvas refactored to single generator (-2,081 lines).
LLM matching in inject disabled by default (deterministic-only unless --enable-llm).
Test suite: 429 tests across 64 files.

Fixed

Ledger write failures no longer crash task updates.
Task list metadata visibility preserved.
Docs: 25 URL redirects for broken/guessable links.
Docs: Corrected wake command flags.

Changelog

On this page