lumberroom-memory 0.7.0

Added

• A fallback for Claude Code's built-in guard, sec-default@builtin, which skips a user-tier plugin's prompt.compose, prompt.section and session.compact hooks on machines with managed settings or a Team or Enterprise login. When a turn ends and prompt.compose never ran, the plugin logs one line, stores a flag in its plugin storage, and attaches the digest and the write rule to a prompt as context once a conversation, again after /clear or a compaction. The flag carries over to later sessions on the machine, so their first prompt gets the copy, and the plugin clears it as soon as prompt.compose runs. On the first eligible prompt the copy waits for the digest fetched with that prompt as query. The built-in memory section and the extraction at compaction stay off under the guard. The README section "Organization-managed Claude Code" covers the admin route that runs every hook. After /clear, which starts the conversation with fresh $.state and no session.start, the plugin reads the stored flag again on the next prompt.

lumberroom-memory 0.6.0

Added

• The digest follows the conversation's first request. The first prompt of 12 characters or more goes to context_bootstrap as query, and the section redraws with the engine's related facts and a profile ordered by relevance. It runs once a conversation, recall on or off, within recallTimeoutMs. A timeout, an error or an engine without query keeps the session-start digest and raises no toast. digestQuery: false turns it off. The refetch runs after the turn goes out, so the first prompt waits for nothing extra and the redrawn section is built to reach the model from the second turn on. No live session has confirmed that yet; docs/spec.md section 14.1 has the check. A refetch that replaces the digest writes one debug line. The system prompt changes once, after that prompt. Needs an engine with decision 0034.

lumberroom-memory 0.5.3

Changed

• lr-review works undated facts: it fills a start date the fact's own text names, through review_queue source undated and the fill_date verdict. Needs a server with the undated source.

lumberroom-memory 0.5.2

Added

• /lr-review works the review queue: dreaming proposals, conflicts and duplicates.

lumberroom-memory 0.5.1

Added

• /lr-setup allows the plugin's own lumberroom calls after you confirm, then loads the digest.
  It adds the missing rules to your user settings.json and keeps every other key.

Changed

• The permissions toast points at /lr-setup instead of a docs link.
• bootstrapTimeoutMs defaults to 5000 ms and accepts no less, so a slow MCP connection still loads the digest.

lumberroom-memory 0.5.0

Removed

• The server, mcpHost and baseUrl options. Install asks for no host: .mcp.json connects to
  https://mcp.lumberroom.cloud/mcp. For a self-hosted engine, register it as an MCP server named
  lumberroom; the plugin falls back to it. The README's Self-hosted engine section has the command.

Changed

• /lr-import uses the lumberroom CLI's login when ingestToken is empty, and posts to the engine
  the CLI points at. Run lumberroom login and pick the Full profile, which carries mayIngest.
• The README shows how to set or change ingestToken, which /config does not list.

lumberroom-memory 0.4.0

Removed

• The classic.SessionStart hook that cut the old lumberroom bootstrap --hook block out of the
  SessionStart context. The plugin directory holds any mod that changes a classic.* result, so
  the plugin now passes that event through untouched. If you still run the old shell hook, remove
  it from ~/.claude/settings.json, or the session gets the digest twice. The digest cache stays:
  it fills the section when the bootstrap fails.

Changed

• The README says which tools the plugin calls itself and when, what the tool.call hook does
  with the calls it sees, the three POST requests /lr-import makes and to which host, and what
  the plugin reads from the conversation and your machine and where it goes.

lumberroom-memory 0.3.2

Added

• skills/lumberroom-memory, the memory rules as a skill: bootstrap at the start, search before
  assuming, write durable facts without asking, and leave review and delete to the user. Since
  engine #96 the MCP server's descriptions give no orders, so in Cowork, where the plugin's hooks
  do not run, the skill is what carries them. The text is the engine's
  client/skills/lumberroom-memory/SKILL.md.

lumberroom-memory 0.3.1

Changed

• mcpHost replaces mcpUrl. It defaults to mcp.lumberroom.cloud, and the bundled MCP server
  connects to https://<mcpHost>/mcp, so a self-hosted engine still needs one setting. .mcp.json
  now spells the scheme out, which the awesome-ai-plugins source scan requires: it rejected a URL
  that was a bare ${user_config.mcpUrl}. If you set mcpUrl, set mcpHost to its host.
• baseUrl defaults to https://<mcpHost>.

Added

• SECURITY.md, with private vulnerability reporting through GitHub.
• Weekly Dependabot updates for the npm development dependencies.
• A README section on each hook, the tools the plugin calls itself and the addresses it contacts.

Fixed

• The importer's byte-order-mark strip wrote U+FEFF as a literal character, which the directory's
  validation holds for review. It is now an escape.

lumberroom-memory 0.3.0

Changed

• The plugin no longer allows its own calls through the permission check. Anthropic's directory does
  not list a mod whose tool.check hook answers allow, so your permission rules now decide. Allow
  context_bootstrap, memory_search and memory_write once, as docs/permissions.md shows.
• extractor drops its options list, which the directory does not accept yet; any value other
  than turn or session-end still reads as off.

Added

• docs/permissions.md, the rules to add for each name lumberroom can run under.
• One toast a session when your permission rules refuse the plugin's calls, linking to that page.