Adds automatic global SKILLS.md bootstrapping for detected agent tool config directories and wires it into memories init/setup, plus a non-blocking global-install postinstall hook; also bumps @memories.sh/cli to 0.7.8 and adds tests/docs. This improves out-of-the-box guidance so agents know when/how to use memories across tools, but introduces filesystem writes in users’ home directories that could be surprising or behave differently across environments despite safeguards to avoid overwriting user-authored SKILLS.md files.

This release refines the CLI’s generate behavior to default to detected integrations so memories generate (and all) won’t emit unrelated outputs, adds a first-class Factory/Droid target, and improves integration detection plus command/docs clarity. It also includes a fix to SQLite FTS trigger logic to prevent SQLITE_CORRUPT_VTAB during hard deletes and refreshes outdated triggers in migrations. Overall it’s mainly a bug-fix/quality patch that makes generated config outputs safer and more predictable for users and integrators.

This change hardens several security- and lifecycle-critical flows: CLI auth now issues expiring one-time auth codes and stores only a SHA-256 token hash (with legacy plaintext fallback), rate limiting tightens client IP trust rules and bounds in-memory limiter growth, and Turso provisioning paths add compensating database cleanup plus retry/backoff metadata. It improves resilience and safety around authentication, billing/membership ordering (Stripe seat billing after membership commit), and account deletion (treat external cleanup failures as retryable 502s, delete auth first, and report pending profile cleanup), and also fixes web build compatibility by removing a Node-only import. Concerns include medium-risk behavioral changes in auth/rate limiting/provisioning that could cause lockouts or mis-rate-limiting if proxy headers/env flags are misconfigured, edge cases around token-hash migration/legacy fallback, and potential for orphaned resources if cleanup paths fail or error handling/order changes differ from existing expectations.

Adds a distributable OpenClaw skill (with a workflows reference runbook) under skills/openclaw and updates the integration docs and OpenClaw resources page to link to it for easier discovery and reuse. This improves onboarding and operational consistency for OpenClaw + memories.sh without changing runtime behavior. Main concerns are that the hardcoded GitHub and default workspace paths could drift or be incorrect for users with custom setups, and the prescribed commands may require ongoing maintenance as the CLI/workspace conventions evolve.

Updates memories generate to default target selection to auto-detected tool integrations (including watch mode), avoiding writing unrelated outputs by default, and adds a first-class factory (Droid) generation target producing .factory/instructions.md with improved tool detection for Factory/Droid and Codex. This changes CLI behavior and generated file set, making output more relevant while expanding supported integrations and clarifying docs that generation exports stored memories to files. Main concerns are detection heuristics potentially missing expected targets or surprising users who relied on the previous “generate everything” default, and the expanded tracked-by-default paths could affect repos that don’t want these files committed.

Adds deterministic tenant mapping source semantics by persisting mapping_source (auto vs override) on sdk_tenant_databases with a backfill migration, updates runtime/management/legacy APIs and dashboard UI to expose the source, and introduces a new management “effective routing” endpoint for tenant routing diagnostics and eligibility decisions. This improves clarity and safety around routing/provisioning behavior (including gating disabling to override mappings) and provides better tooling for debugging routing outcomes, but it carries risk around the DB migration/backfill and any misclassification of legacy rows (metadata-based fallback) potentially affecting provisioning/disable flows and routing decisions.

This change makes tenant database routing “auto” by default and introduces a new authenticated management endpoint (/api/sdk/v1/management/tenant-overrides) for listing/upserting/disabling explicit tenant→DB overrides, updating the core SDK client, dashboard UX, and docs to use it while keeping legacy routes as aliases. It reduces setup friction for SaaS users and enables headless server-to-server override management via bearer API keys, but it also expands a high-risk surface that can provision/attach/disable tenant databases, so auth/rate-limit/billing gating mistakes or provisioning failures could lead to unintended DB operations, spend, or tenant routing regressions. Additional concerns include backward-compatibility/transition issues between legacy endpoints and the new envelope behavior, plus operational fragility around Turso provisioning/verification timing and error handling.

This updates the Tenant Routing UI to default to an auto-routing summary (API key health, fixed “Automatic” mode, and mapping counts) and moves tenant override create/attach/disable controls into a collapsible “Advanced: Tenant Overrides” section, while adjusting gating so Growth is required only for override-management actions and mapping inspection/refresh depends only on having a valid API key. The project impact is a clearer, more mobile-friendly routing experience with fewer unnecessary plan restrictions, and it restores visibility into mappings for non-Growth users when they have a valid key. Potential concerns are user confusion/discoverability since override actions are now hidden behind an advanced toggle and can be disabled based on plan/key state, plus behavior changes around when mappings are cleared/hidden could surprise users relying on previous gating.

This change adds a new OpenClaw section to the web app, including a dedicated landing page, a resources hub, and an agent-oriented llms.txt runbook with copyable CLI setup/refresh workflows. It expands the site’s navigation and sitemap to surface and index these new routes while introducing new OpenClaw-specific UI components (hero background/illustration and interactive command sequence). Main concerns are the large volume of new frontend code/content increasing maintenance and bundle weight, and potential UX/linking edge cases from the updated navbar anchor resolution across routes (plus keeping the runbook commands accurate over time).

@memories.sh/cli 0.7.7 is a patch release that refines memories generate so it defaults to only the integrations it detects, avoiding writing unrelated outputs (e.g., Copilot/Gemini) unless present. It adds a first-class factory target for Droid (.factory/instructions.md), improves detection for Factory/Droid and Codex CLI installs, and clarifies messaging/docs that generate exports stored memories to config files rather than creating new database memories—making output safer and clearer for users and integrators.

@memories.sh/core adds first-class, typed management APIs for keys and tenant mappings (keys: get/create/revoke; tenants: list/upsert/disable) and supports tenant-scoped SDK usage via a new tenantId option that’s propagated through tool calls. This is primarily a feature release that reduces the need for raw HTTP calls, improves typed error/response handling, and makes multi-tenant integrations clearer for users and SDK integrators.

@memories.sh/cli 0.7.6 includes both a minor feature update and a patch fix: setup/ingestion now pulls in a broader set of project skills by default (with a --skip-skill-ingest opt-out), and the SQLite FTS trigger logic was corrected to prevent SQLITE_CORRUPT_VTAB errors when hard-deleting soft-deleted memories while auto-refreshing outdated triggers during migrations. This improves out-of-the-box agent skill coverage and makes database operations/migrations more reliable for users and integrators.

This release includes a patch fix to the SQLite full‑text search trigger logic, preventing SQLITE_CORRUPT_VTAB errors when hard-deleting previously soft-deleted memories and auto-refreshing outdated trigger definitions during migrations. It also improves agent setup by ingesting existing project skills by default (with a new --skip-skill-ingest opt-out) and expanding memories ingest skills to scan more supported skill directories. Overall, it’s a mix of bug fixes and usability/feature improvements that should make upgrades safer and setup/integration smoother.

This release adds first-class, typed management APIs in the SDK for managing keys and tenant mappings (e.g., keys get/create/revoke; tenants list/upsert/disable), plus new helpers/types for AI SDK users. It reduces the need for raw HTTP calls and makes tenant-scoped integrations clearer and safer via improved typing and scoped client/tool behavior—primarily a feature release.

@memories.sh/ai-sdk 0.3.1 rolls forward the recent 0.3.x updates: it adds first-class, typed management APIs for keys and tenant mappings (via new helpers like memoriesManagement(), managementKeys(), and managementTenants()) and pulls in the corresponding @memories.sh/core update. This is primarily a feature release with dependency updates, making it easier for users/integrators to manage tenants and API keys directly through the SDK instead of raw HTTP calls.

This release includes a patch fix to the SQLite FTS trigger logic to prevent SQLITE_CORRUPT_VTAB errors when hard-deleting soft-deleted memories, and it auto-refreshes outdated trigger definitions during migrations. It also improves agent setup by ingesting existing project skills by default (with a new --skip-skill-ingest opt-out) and expands memories ingest skills to scan more skill directories. Overall it’s both a bug-fix and feature update, making setups smoother and data/migrations more reliable for users and integrators.

This CLI release includes a patch that fixes SQLite FTS trigger logic to prevent SQLITE_CORRUPT_VTAB errors when hard-deleting previously soft-deleted memories, and it refreshes outdated trigger definitions during migrations. It also improves agent setup by ingesting existing project skills by default (with a new --skip-skill-ingest opt-out) and expands memories ingest skills to scan more supported skill directories—making setups smoother and integrations more reliable. Overall it’s both a bug-fix and feature/usability update.

@memories.sh/cli 0.7.2 is a patch bug-fix release that updates the full-text search (FTS) trigger logic to prevent SQLITE_CORRUPT_VTAB errors when hard-deleting previously soft-deleted memories. Migrations now also auto-refresh outdated trigger definitions, improving database stability and reducing maintenance issues for users and integrators.

This release improves the CLI’s default skill ingestion for agent setups: memories setup now imports existing project skills automatically (with a new --skip-skill-ingest opt-out), and memories ingest skills scans all supported skill directories including .codex/skills. It’s a feature-focused update that makes initial setup and integrations more complete by pulling in more skills by default, reducing manual ingestion work for users.

@memories.sh/cli 0.7.0 improves skill ingestion during agent setup: memories setup now imports existing project skills by default, and memories ingest skills scans all supported skill directories (including .codex/skills). Users and integrators should see more complete, automatic skill coverage out of the box, with a new --skip-skill-ingest flag to opt out—this is a feature-focused minor release.