Highlights

• Upgraded mu's pi runtime dependencies to 0.56.2 across agent and CLI packages.
• Released synchronized workspace version 26.3.7 for all publishable mu packages.

User-visible changes (features/fixes)

• @femtomc/mu-agent now depends on:
  • @mariozechner/pi-agent-core@^0.56.2
  • @mariozechner/pi-ai@^0.56.2
  • @mariozechner/pi-coding-agent@^0.56.2
  • @mariozechner/pi-tui@^0.56.2
• @femtomc/mu (CLI) now depends on:
  • @mariozechner/pi-ai@^0.56.2
  • @mariozechner/pi-coding-agent@^0.56.2

Operational/tooling changes

• Refreshed bun.lock to capture [email protected] and resulting transitive dependency updates.
• Bumped release version to 26.3.7 in root and all publishable manifests.
• Validation completed via bun run check before release.

Docs/branding updates

• Synced periodic logo version marker from 26.3.6 to 26.3.7 in assets/mu-periodic-logo.svg.

Highlights

• Enforced progressive-disclosure defaults across high-volume CLI read surfaces to reduce accidental context blowups in agent/operator sessions.
• Added explicit escape hatches (--json, --raw, --all) where full payloads are required.

User-visible changes

• mu replay
  • bounded preview by default (tail preview)
  • full dump now requires explicit --all --raw
  • added --raw for exact tail lines
• mu store tail
  • compact preview by default
  • added --raw for exact row output
  • reduced max --limit to 500
• mu issues list and mu issues ready
  • default bounded output (--limit 20)
  • explicit unbounded mode via --all
• mu forum read and mu forum topics
  • bounded defaults and max caps
  • forum topics supports explicit --all
• mu issues get
  • compact mode now shows body preview instead of full body dump
• mu control identities
  • compact table by default, full records via --json
• mu control operator models / mu control operator thinking <provider>
  • compact summaries by default
  • full per-model detail via --verbose/--debug (or --json)

Operational/tooling changes

• Added CLI regression tests covering bounded defaults and explicit full-output pathways.
• Updated command help and guide text to steer toward compact-first workflows.

Docs/branding

• Synced periodic logo embedded version to 26.3.6.
• Updated CLI docs/examples to reflect progressive-disclosure defaults.

Compare

• v26.3.5...v26.3.6

Highlights

• Renamed the starter writing skill from writing to technical-writing.
• Replaced the old generic writing guidance with a stronger systems/technical writing playbook focused on argument quality, design rationale, and evidence-backed claims.
• Added migration cleanup so legacy top-level writing skill directories are pruned during starter-skill bootstrap.

User-visible changes

• New starter skill path and name:
  • skills/technical-writing/SKILL.md
• Removed legacy starter skill file:
  • skills/writing/SKILL.md
• Updated operator-facing references so skill handoff now points to technical-writing.
• Expanded technical-writing guidance to include:
  • problem/insight/evidence/limits argument spine
  • phase-based paper workflow (outline → sentence skeleton → polish → feedback)
  • checklists for intro/related work/design/evaluation/discussion
  • common systems-paper failure modes and review passes

Operational and tooling changes

• Starter skill bootstrap now hard-prunes legacy top-level writing directory entries to avoid stale local installations.
• CLI starter-skill bootstrap tests updated to assert technical-writing seeding and prompt injection.

Docs and branding

• @femtomc/mu-agent README starter-skill list updated to technical-writing.
• Versioned logo asset synced to 26.3.5.

Validation

• bun run check passed before release.

Highlights

• Added durable programmable-UI persistence so mu_ui state survives session resume/switch/fork/tree navigation.
• Introduced a new control-plane UiDoc state journal (ui_docs_state.jsonl) with scope/revision semantics for async multi-actor coordination.
• Added a new CLI surface: mu control ui-state for listing/inspecting persisted UiDoc scope snapshots.

User-visible changes

• mu_ui extension now stores snapshot state via session custom entries (mu-ui-state/v1) and restores latest valid snapshot on session lifecycle events.
• New control command group:
  • mu control ui-state
  • mu control ui-state list ...
  • mu control ui-state get ...
• CLI README/help updated with ui-state command usage.

Operational / tooling changes

• Control-plane runtime now loads and serves UiDocsStateStore from:
  • <workspace>/control-plane/ui_docs_state.jsonl
• Command pipeline now journals normalized ui_docs from operator responses into scope-specific state records.
• Added test coverage for:
  • ui extension persistence/restore
  • ui docs state store behavior
  • pipeline-to-store integration
  • CLI control ui-state behavior

Docs / branding updates

• Updated @femtomc/mu-agent and @femtomc/mu-control-plane READMEs for new persistence behavior and store semantics.
• Synced periodic logo version marker to 26.3.4.

Highlights

• Hardened autonomous operator wake failure handling so provider-side failures are classified and surfaced explicitly instead of collapsing into generic operator_empty_response outcomes.
• Heartbeat wake decisions now treat fallback-coded operator replies (for example provider quota/context failures) as failed wake outcomes, which correctly marks heartbeat ticks as failed instead of ok.
• Added focused regression coverage across agent runtime/backend and server wake/tick telemetry paths.

User-visible changes

• New operator fallback/error codes are now emitted when upstream provider failures are detected from assistant error payloads:
  • operator_provider_usage_limit
  • operator_provider_context_length_exceeded
• Fallback safe-response messages now include these provider failure codes and clearer explanations.
• Heartbeat-triggered wakes that return fallback-coded provider errors now produce failure semantics in scheduler state (last_result: "failed" with explicit last_error) rather than reporting successful dispatch.

Operational/tooling changes

• Strengthened wake decision classification in mu-server so non-busy fallback codes are tracked as fallback failures while preserving no-spam/coalescing behavior.
• Regression tests added/updated:
  • packages/agent/test/operator_runtime.test.ts
  • packages/agent/test/operator_backend.test.ts
  • packages/server/test/server.test.ts

Docs/branding

• Versioned periodic logo asset synchronized to 26.3.3.

Compare

• Full diff: v26.3.2...v26.3.3

Highlights

• Heartbeat programs can now route autonomous turns through per-program provider/model/thinking overrides.
• Heartbeat programs can now pin autonomous turns to explicit context checkpoints (session_id, session_file, session_dir).
• Autonomous operator runtime/backend now honors wake-scoped session/model overrides with profile-aware session recreation.

User-visible changes

Heartbeats: per-program model routing

You can now configure heartbeat-local routing without changing workspace-global operator defaults.

• operator_provider
• operator_model
• operator_thinking

CLI additions:

• mu heartbeats create --provider <id> --model <id> [--thinking <level>]
• mu heartbeats update --provider <id> --model <id> [--thinking <level>]

Validation guardrails:

• --provider and --model must be provided together.
• --thinking requires provider+model.

Heartbeats: context checkpoints

Heartbeat programs can now target stable autonomous session checkpoints:

• context_session_id
• context_session_file
• context_session_dir

CLI additions:

• mu heartbeats create --session-id <id> [--session-file <path>] [--session-dir <path>]
• mu heartbeats update --session-id <id> [--session-file <path>] [--session-dir <path>]

Behavioral details:

• Wake metadata carries routing/checkpoint fields through autonomous ingress.
• For heartbeat wakes without explicit context_session_id, server defaults to heartbeat-program:<program_id>.

Output/rendering

Compact heartbeat rendering now includes configured model/checkpoint context when present.

Operational/tooling changes

• API surface updated for /api/heartbeats/create and /api/heartbeats/update to accept routing/checkpoint fields.
• Server wake dispatch path now includes routing/checkpoint metadata in wake payloads and autonomous ingress metadata.
• Regression coverage expanded across server/agent/cli to verify end-to-end routing/checkpoint behavior and validation.

Docs/branding updates

• CLI README updated with heartbeat model-routing and checkpoint examples.
• Versioned periodic logo synced to 26.3.2.

Compare

• v26.3.1...v26.3.2

Highlights

• Hardened heartbeat/cron wake behavior to reduce race conditions and confusing timeout/busy outcomes.
• Added heartbeat ownership guardrails for mu issues mutations, with explicit override when human intervention is intended.
• Fixed memory index rebuild failures caused by duplicate item_id collisions in control-plane sources.

User-visible changes

• Scheduling / wake reliability:

  • Manual trigger commands now use a longer request timeout:
    • mu heartbeats trigger ...
    • mu cron trigger ...
  • Heartbeat program registry now coalesces concurrent in-flight ticks per program.
  • operator_busy wake-turn outcomes are treated as coalesced wake results instead of hard failures.

• Heartbeat ownership guardrails for issue DAG mutation:

  • mu issues mutation commands now block changes targeting heartbeat-managed roots unless explicitly overridden.
  • New override flag:
    • --allow-heartbeat-managed
  • Applies across create/update/claim/open/close/dep/undep mutation paths.
  • Autonomous heartbeat-owned turns are allowed when wake context matches the owning heartbeat program.

• Memory index stability:

  • Rebuild path now prevents duplicate memory_items.item_id collisions for control-plane-derived rows.
  • Added defensive unique-id normalization before index writes.

Operational / tooling changes

• Expanded regression coverage for:
  • concurrent heartbeat trigger coalescing,
  • heartbeat ownership mutation blocking and autonomous ownership allowance,
  • memory index rebuild with repeated control-plane IDs.
• Release manifests updated to 26.3.1 across publishable packages.

Docs / branding

• CLI docs updated with heartbeat-ownership override guidance.
• Versioned logo asset synchronized to 26.3.1.

Compare

• v26.2.120...v26.3.1

Highlights

• CLI startup and command dispatch paths were tightened to reduce cold-start overhead for common invocations (mu, mu serve, help paths).
• mu serve startup-related extension refreshes continue to run in a non-blocking/coalesced mode, improving perceived attach responsiveness.

User-visible changes

• Faster command startup for lightweight and help invocations:
  • command handlers are now loaded lazily per command path rather than eagerly as a full CLI graph.
  • help requests avoid store/bootstrap setup.
  • runtime context selection is stricter so only store-heavy commands materialize full issue/forum/event stores.
• mu serve / operator-session orchestration now benefits from reduced up-front command-load cost.
• Running-server detection remains behaviorally compatible, with a shorter health probe timeout for stale/unresponsive server discovery.

Operational and tooling changes

• @femtomc/mu* workspace packages bumped to 26.2.120.
• Lockfile refreshed for release packaging.
• Versioned logo metadata synced to 26.2.120.

Docs / branding updates

• No user-facing docs changes in this release.
• Periodic logo version stamp updated as part of release synchronization.

Highlights

• Improved responsiveness across both CLI startup/store initialization and terminal UI refresh paths.
• Reduced redundant UI status/widget writes and avoided startup blocking on branding status refresh.

User-visible changes

• Terminal mu_ui refresh path now skips duplicate status/widget updates when visible content is unchanged.
• Branding status refresh no longer blocks initial session start/switch rendering and is coalesced when polling overlaps.
• Branding status fetch timeout reduced to improve perceived responsiveness under slow server conditions.

Operational / tooling changes

• CLI workspace context setup now parallelizes module/repo-root resolution.
• Bundled starter-skill bootstrap now avoids unnecessary file reads for existing files and caches bundled skill metadata lookups in-process.
• Release cut for v26.2.119 with synchronized manifests, logo version, and lockfile refresh.

Docs / branding updates

• Versioned logo text updated to 26.2.119.

Compare

• v26.2.118...v26.2.119

Highlights

• Added a tmux-friendly programmable-UI interaction shortcut fallback in terminal operator sessions.
• Preserved existing ctrl+shift+u behavior while adding alt+u for environments that drop Shift on Ctrl+letter chords.

User-visible changes

• Terminal programmable-UI modal can now be opened with either:
  • ctrl+shift+u (primary)
  • alt+u (fallback)
• Auto-prompt notification copy now references both shortcuts.
• Updated UI extension tests to validate both shortcut registrations.

Operational / tooling changes

• Release cut for v26.2.118.
• Version bump across publishable manifests and lockfile refresh.

Docs / branding updates

• Updated docs/mu-ui.md to document the fallback shortcut and tmux/remote-terminal caveat.
• Updated packages/agent/README.md shortcut guidance to include fallback behavior.
• Synced versioned logo text to 26.2.118.

Compare

• v26.2.117...v26.2.118