v13.41.0
thedotmack/claude-memv13.41.0Oct 17, 2026by thedotmack
Upgrading?
AI Summary
This release introduces new high-level workflow skills for building and releasing projects, alongside enhanced privacy-preserving telemetry for tracking work-state usage. It focuses on diagnostic signals rather than compliance scores and ensures no sensitive data is exposed in the new rollups. Additionally, the release restores the `childhood` skill for post-release verification.
Key Highlights
- New `build`, `release`, and `build-and-release` skills that delegate to existing workflows.
- Enhanced privacy-preserving PostHog rollups for work-state telemetry (reads/writes, outcomes, handoffs).
- Detailed measurement of instruction handoffs, excluding failed outputs and untimestamped captures.
- Restoration of the `childhood` skill for post-release verification.
- Strict privacy controls ensuring no raw prompt text, IDs, or project paths are sent.
New Features
- New `build` skill for delegating coding sessions.
- New `release` skill for publishing versions.
- New `build-and-release` skill for merging PRs and updating main.
- PostHog rollups for work-state telemetry distinguishing reads and writes.
- MCP execution receipts covering failures before worker response.
- Instruction handoff tracking with specific success criteria.
Full Release Notes
# Build, release, and work-state visibility ## Workflow skills - `build` delegates a coding session through `make-plan`, `do`, and `babysit`. - `release` delegates publication through `version-bump` and `childhood`. - `build-and-release` runs both, merging the ready PR before continuing from updated main. - Restores the existing `childhood` skill so post-release verification is included. The new wrappers are deliberately short. They reuse the existing workflows. ## Work-state telemetry Adds privacy-preserving PostHog rollups for work-state usage by release, host platform, and recorded agent role. The rollups distinguish captured reads and writes, outcomes, native to-do use, instruction handoffs, and missing attribution. MCP execution receipts also cover failures before a worker responds. They are a separate observation source and must not be added to captured-call totals. Instruction handoffs are measured after the platform adapter formats the model-bound output and stdout succeeds. Cached context is included; discarded context and failed output are not counted as successful handoffs. Delayed or untimestamped captures cannot inherit a later instruction handoff. These are diagnostic signals, not a compliance score. Returning instructions does not prove that a host accepted them or an agent followed them, and an ordinary question may correctly require no write. No task titles, prompt text, project paths, or raw session/agent IDs are sent. Existing telemetry opt-out remains in effect. This release does not backfill earlier usage. Local replay deduplication lasts 30 days; older events are dropped. Public/private turn boundaries remain content-free and prevent private activity from inheriting an earlier public prompt. Hosts without verified original timestamps remain outside captured prompt/tool totals.
Related Repositories
The AI that really does things. Any OS. Any Platform. The lobster way. 🦞
Interactive roadmaps, guides and other educational content to help developers grow in their careers.
Reverse engineer anything with agents, from app behavior down to native binaries.
The open source coding agent.
Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.