v0.33.0

What's Changed

• docs: archive deprecated designs under design/_archived, move qa/rca/reviews under docs/dev by @danieljhkim in #5925
• Design an email channel for human decisions on held Orbit work by @orbit-agent-01 in #5926
• [auto-task] Validate the oldest workspace documentation by @orbit-agent-01 in #5931
• Keep focused phone navigation visible through rail resizing by @orbit-agent-01 in #5927
• Reconcile dashboard host-block focus ring validation by @orbit-agent-01 in #5928
• Route task-pilot's Source reader and pilot reservation through TaskAutomationPort by @orbit-agent-01 in #5933
• Document the Claude Code Orbit band, board and ship pane on the website by @orbit-agent-01 in #5930
• Document the Orbit plugin installation, grants and workspace lifecycle on the website by @orbit-agent-01 in #5929
• [friction-curation] Repair stale dashboard context browser fixture expectations by @orbit-agent-01 in #5934
• [friction-curation] Retire the hidden --pretty error alias of orbit tool run by @orbit-agent-01 in #5935
• Settings topic drawer corrupts array values that contain commas and writes 0 for a cleared integer by @orbit-agent-01 in #5938
• [friction-curation] Let delivery commit a new crate guide AGENTS.md symlink by @orbit-agent-01 in #5937
• [friction-curation] Tell implementers to omit regression_from when no culprit exists by @danieljhkim in #5936
• Settings file view reads raw TOML through an unvalidated path by @orbit-agent-01 in #5940
• Settings topic pages: restyle the config rows and drop the dead cfx/cwf/cgf CSS by @orbit-agent-01 in #5941
• Document first use of the Orbit desktop app and fleet tray on the website by @orbit-agent-01 in #5939
• [delivery-code-review] Claude Code guide incorrectly says every failed preflight check prevents Launch by @orbit-agent-01 in #5944
• [friction-curation] Route a task whose criterion runs an OS-gated test to that OS by @orbit-agent-01 in #5942
• Configurable PR, commit and branch conventions per workspace by @orbit-agent-01 in #5943
• All keys raw TOML can show the other file after overlapping loads by @danieljhkim in #5946
• Desktop app guide gives wrong log locations and tray status text by @orbit-agent-01 in #5947
• [friction-curation] Give an escalated publication with a green PR a landing route by @orbit-agent-01 in #5948
• Settings drawer CLI preview renders numeric settings as TOML strings by @danieljhkim in #5950
• Preserve the task delivery marker when truncating templated commit subjects by @orbit-agent-01 in #5949
• Record the 2026-10-25 CPU bias-correction replay in §4.2 by @orbit-agent-01 in #5945
• Escalated requeue treats a rollup node with non-boolean isRequired as not required by @orbit-agent-01 in #5951
• Templated commit truncation exceeds 72 characters for a 71-character marker suffix by @orbit-agent-01 in #5952
• Keep delivery PR identity when a repository body template omits the task ID by @orbit-agent-01 in #5953
• Add explore-first tag instruction for the explorer trial by @danieljhkim in #5955
• Number-row CLI previews emit out-of-range TOML integer literals for large finite floats by @orbit-agent-01 in #5954
• Document operator command and job tools over MCP by @danieljhkim in #5957
• Complete activity and job authoring reference by @orbit-agent-01 in #5960
• Document audit and log investigation from the terminal by @orbit-agent-01 in #5962
• Document credentials for scheduled and macOS Claude runs by @orbit-agent-01 in #5958
• Document executor YAML overrides and host constraints by @orbit-agent-01 in #5956
• Document tag instructions in the orbit-setup skill by @orbit-agent-01 in #5959
• Document the dashboard Plugins view by @orbit-agent-01 in #5964
• Track Markdown fence delimiters before enforcing required PR sections by @orbit-agent-01 in #5961
• Document manual task file reservations by @orbit-agent-01 in #5965
• Document attaching and retrieving task evidence by @orbit-agent-01 in #5966
• Explain dashboard analytics and resource profiles by @orbit-agent-01 in #5967
• Activity YAML reference says a target step's timeout_seconds bounds dispatch, but the executor never reads it by @orbit-agent-01 in #5969
• Correct executor fallback docs: an OS-mismatched sandbox is always refused by @orbit-agent-01 in #5972
• Document plugin manifest authoring and conformance by @danieljhkim in #5968
• [auto-task] Validate the oldest workspace documentation by @orbit-agent-01 in #5973
• Activity YAML reference says final_recovery_activity only resumes or settles without the failure hook by @orbit-agent-01 in #5970
• code-review auto-task reviews the 2 least recently reviewed modules by @orbit-agent-01 in #5971
• Document provider identity, crew resolution order and per-provider CLI setup in the orbit-setup skill by @orbit-agent-01 in #5974
• Align the new file-reservation guide with the enforced four-hour TTL limit by @orbit-agent-01 in #5975
• Plugin manifest reference misstates who may call a mutating tool by @orbit-agent-01 in #5977
• Host list spec says every reachable field is null with --no-probe by @orbit-agent-01 in #5979
• Make run-failure filing use valid tags and portable routing by @danieljhkim in #5976
• Document the delivery, admission and housekeeping settings the orbit-setup skill never reaches by @danieljhkim in #5978
• Document auto-task YAML and non-cron routine triggers by @orbit-agent-01 in #5980
• chore: Trim the default config template to the keys it sets by @danieljhkim in #5981
• Document reporting and triaging friction by @orbit-agent-01 in #5982
• Correct the seeded-routine count and document the crew-performance refresh routine in the orbit-setup skill by @orbit-agent-01 in #5984
• Group agent explore and invoke modules by @orbit-agent-01 in #5985
• Move runtime.rs to runtime/mod.rs by @orbit-agent-01 in #5987
• [delivery-code-review] Provider setup skill calls the doctor BWRAP column the CLI launcher by @orbit-agent-01 in #5986
• Rail sub-items: keep labels like "Auto-tasks" on one line by @orbit-agent-01 in #5989
• Leave no-diff-expected tasks out of the implementer scorecard by @orbit-agent-01 in #5991
• Document full state backups and task archive transfers by @orbit-agent-01 in #5988
• Document workspace removal, teardown and source-remote rebinding in the orbit-setup skill by @orbit-agent-01 in #5992
• Let workspaces map a task tag to an agent review skip by @orbit-agent-01 in #5993
• Make full-QA roll-up evidence reads require the owner lane by @orbit-agent-01 in #5990
• Window length presets: drop 15m/30m, offer 1h 2h 4h 8h 12h 24h by @orbit-agent-01 in #5995
• feat: Seed current built-in crews and write every crew route explicitly by @danieljhkim in #5983
• Recovery agents treat two per-command permission denials as "no shell access" and skip validation and friction filing by @orbit-agent-01 in #5996
• Keep QA sweep baseline extraction under run-scoped scratch by @orbit-agent-01 in #5998
• Map shipped auto-task tags to review skips in default config by @orbit-agent-01 in #6000
• Remove unsafe live SQLite file-copy backup advice by @orbit-agent-01 in #6001
• Rail foot: one-line refresh status and no divider above the host block by @orbit-agent-01 in #5994
• Task pilot holds promotion on a prerequisite already recorded as a blocked_by dependency by @orbit-agent-01 in #5999
• Keep mapped review-skip delivery previews read-only by @orbit-agent-01 in #6003
• Environment default crew still looks up removed antigravity and muse names by @orbit-agent-01 in #6007
• [delivery-code-review] Module cleanup task cannot be filed when grouped findings share a culprit by @orbit-agent-01 in #6004
• [friction-curation] Refuse workspace init that resolves to a live workspace by @orbit-agent-01 in #6006
• Dashboard: "Run pilot" action on proposed tasks submits an explicit task_pilot_pipeline run by @orbit-agent-01 in #6002
• Init crew docs still describe the pre-rename built-in registry by @orbit-agent-01 in #6009
• Run mapped review-skip guards in the candidate checkout by @orbit-agent-01 in #6005
• Retire the Cursor, Pi and OpenCode provider lanes by @orbit-agent-01 in #6008
• Document managed workspace maintenance and removal by @orbit-agent-01 in #5997
• [code-review] Reconcile runbooks with owner search, push-only CodeQL and portable header checks by @orbit-agent-01 in #6011
• [code-review] Credential reference still treats execution.env.pass as the only operator allowlist by @orbit-agent-01 in #6012
• Document the embedded MCP Apps control center by @orbit-agent-01 in #6010
• Split RuntimeHost impl out of host.rs by @orbit-agent-01 in #6013
• Split the RuntimeHost trait body by @orbit-agent-01 in #6014
• Preserve explicit --workspace selection in config-write refusal checks by @orbit-agent-01 in #6015
• Split orbit-config settings table.rs by @orbit-agent-01 in #6016
• Retired Cursor/Pi/OpenCode lanes still offered in current website, SECURITY and skill docs by @orbit-agent-01 in #6017
• Attribute regression_from in QA and full-review findings by @orbit-agent-01 in #6019
• Offline decision-tree report suggesting crew performance types by @orbit-agent-01 in #6020
• MCP Apps guide rejects a valid control-center open by @orbit-agent-01 in #6023
• Dashboard: "Fix PR" action on review tasks starts a single-task out-of-run auto-fix by @orbit-agent-01 in #6021
• MCP Apps guide says a task edit changes one field at a time by @orbit-agent-01 in #6024
• Fix Muse provider usage extraction for Muse Code 1.4.13 run.usage frame by @orbit-agent-01 in #6022
• Ship a daily comment-hygiene auto-task default by @orbit-agent-01 in #6025
• MCP Apps guide says automation jobs show a next scheduled slot by @orbit-agent-01 in #6027
• docs: Scope a conflict budget for footprint admission by @danieljhkim in #6028
• [delivery-code-review] Discovery TOML suggestions do not preserve tree-leaf predicates by @orbit-agent-01 in #6026
• fix: Extract Muse usage windows from the 1.4.13 run.usage frame [ORB-18154] by @danieljhkim in #6018
• Fix the per-crate Clippy guide command for binary-only orbit-cli by @orbit-agent-01 in #6029
• Dashboard: flag review tasks whose delivery run failed or whose PR is red/conflicting by @orbit-agent-01 in #6030
• Ship architecture-audit as a portable default auto-task by @orbit-agent-01 in #6031
• Make admitted branch templates identify their actual task without prefix collisions by @orbit-agent-01 in #5963
• Per-path footprint lock modes (exclusive/shared) with regex patterns in workspace config by @orbit-agent-01 in #6032
• [friction-curation] Name the sender of a SIGTERM that kills a pipeline worker by @orbit-agent-01 in #6033
• [delivery-code-review] Discovery emits the reserved other type name as ready-to-paste configuration by @orbit-agent-01 in #6034
• Refresh expired Fix PR readiness before reusing an unchanged task detail by @orbit-agent-01 in #6035
• Add a dependency-freshness default auto-task by @orbit-agent-01 in #6036
• [delivery-code-review] Prepare the comment-hygiene footprint before admission by @orbit-agent-01 in #6040
• Ship path-risk-review as a default auto-task by @orbit-agent-01 in #6044
• [code-review] Align setup skill with the code-review module rotation by @orbit-agent-01 in #6047
• Clear superseded auto-fix warnings after successful recovery by @orbit-agent-01 in #6049
• [qa-sweep] crew-type-discovery CLI dumps a Python traceback for a missing or malformed export by @orbit-agent-01 in #6048
• Recognize branch ownership when an external ref contains / by @danieljhkim in #6050
• Ship structure-sweep as a portable default auto-task by @orbit-agent-01 in #6037
• Align implementer regression output with its null guidance by @danieljhkim in #6052
• Keep clean-handoff report schemas in the execution reference by @danieljhkim in #6055
• Decision channel slice 1: record and raise decision requests by @danieljhkim in #6051
• Add an auto-task-yield-review default auto-task by @orbit-agent-01 in #6054
• Add a suppression-ratchet default auto-task by @orbit-agent-01 in #6043
• Reference tool argument guidance from the implementer prompt by @danieljhkim in #6058
• Conflict recovery at complete_pr is refused after a before-landing push by @orbit-agent-01 in #6061
• Decision requests remain open after the held task leaves its block episode by @orbit-agent-01 in #6062
• [auto-task] Website accuracy — validate the oldest orbit-cli.com pages against the current release by @orbit-agent-01 in #6064
• Yield review: discover incoming task relations correctly by @orbit-agent-01 in #6063
• github.run.list: workflow miss is an execution failure by @orbit-agent-01 in #6059
• A conflict-recovered head never reaches the PR, so completion fails review_gate_stale then delivery_evidence_stale by @orbit-agent-01 in #6065
• Yield review: exhaust the full 30-day task population by @orbit-agent-01 in #6069
• Document the required --by-type switch for scorecard --shadow by @orbit-agent-01 in #6072
• Dashboard: "Triage" action runs escalation_triage_pipeline for a held proposed task or an escalated blocked task by @danieljhkim in #6070
• Skill tells agents to poll an operator-only run tool by @orbit-agent-01 in #6071
• Bring the cross-workspace dashboard aggregates /api/tasks/all and /api/fleet/drains under 500 ms p95 with no tail over 1 s by @orbit-agent-01 in #6068
• Add an escape-to-guard default auto-task for recurring regressions by @orbit-agent-01 in #6053
• Decision channel slice 4: DecisionRequested state trigger by @orbit-agent-01 in #6073
• Repaint Fix PR readiness for tasks opened outside the current task list by @orbit-agent-01 in #6042
• Auto-fix rewind still picks the wrong pr_complete round for production-shaped checkpoints (failed round is never recorded) by @orbit-agent-01 in #6074
• [code-review] Align the run-triage eval with the orchestration skill by @orbit-agent-01 in #6075
• [code-review] Reject unknown bundle IDs when the allowed ID set is empty by @orbit-agent-01 in #6076
• Ship code-health-sweep as a portable default auto-task by @orbit-agent-01 in #6057
• Keep delivery warnings visible beyond 500 unrelated runs by @danieljhkim in #6077
• [qa-sweep] github.run.list reports an unknown workflow file name or numeric id as an execution failure, not invalid input by @orbit-agent-01 in #6079
• A reviewer's reverted-fix reproduction blocks landing as validation_contradicted by @orbit-agent-01 in #6080
• Ship ci-budget-watch as a portable default auto-task by @orbit-agent-01 in #6078
• Gate the Linux signal-sender regression test on Linux by @orbit-agent-01 in #6082
• Add a flaky-test-hunt default auto-task by @danieljhkim in #6084
• Cut /api/routines memo-miss compute: batch routine reads and share one task snapshot by @danieljhkim in #6085
• Decision channel slice 2: list and answer decision requests from CLI, MCP and dashboard by @orbit-agent-01 in #6081
• Runbook: recover a delivery held at landing_review_gate_settle by @orbit-agent-01 in #6087
• Ship agent-guidance-audit as a default auto-task by @danieljhkim in #6086
• Apply the chosen triage replacement before returning work to backlog by @orbit-agent-01 in #6088
• Dashboard: a "Why stuck" card for blocked and held tasks by @danieljhkim in #6089
• Show decision option labels and consequences before dashboard answers by @orbit-agent-01 in #6091
• Why stuck card mislabels an abandoned landing review as verdict incomplete, by @orbit-agent-01 in #6092
• Why stuck card mislabels a blocked landing review as abandoned when its escalation quotes review_abandoned: by @orbit-agent-01 in #6093
• Decision channel slice 3: delegated-human channels and governed orbit.decision.submit by @orbit-agent-01 in #6090
• [auto-task] Check the oldest code comments for staleness by @orbit-agent-01 in #6094
• Enforce operator authority on the direct decision-answer CLI by @danieljhkim in #6096
• Bring /api/routines memo-miss under 200 ms and orbit routine list under 250 ms p95 by @orbit-agent-01 in #6095
• Decision channel sender check refuses operator omission yet accepts an empty plugin sender by @orbit-agent-01 in #6098
• Do not consume pilot decision requests when the selected verdict leaves the hold unresolved by @danieljhkim in #6097
• Escalation triage refuses tasks whose required_tools include tools the triager may not use by @orbit-agent-01 in #6099
• Raise another pilot decision when resolving one typed hold reveals another in the same assessment by @orbit-agent-01 in #6100
• Test fixtures leak a detached auto-drain worker that outlives the test run by @orbit-agent-01 in #6103
• Seed before-landing review into PR-route workspaces at init by @danieljhkim in #6101
• Dashboard: requeue and settle-merged actions for blocked and held tasks by @danieljhkim in #6104
• candidate_path_authority_contracts .git component case fails on macOS: Git hides notes/.GIT, so commit reports nothing to commit by @danieljhkim in #6106
• Blocked-task triage non-Git fixture reaches the enclosing checkout's .git when TMPDIR is nested in a Git checkout by @orbit-agent-01 in #6105
• Dashboard: one-click operator records for pilot-admission holds and owed post-deploy criteria by @orbit-agent-01 in #6107
• qa-full-sweep harness loses its whole report when a fixture writes into its temp dir during cleanup by @orbit-agent-01 in #6108
• [auto-task] Validate the oldest workspace documentation by @orbit-agent-01 in #6110
• Show pilot admission actions for held backlog tasks by @orbit-agent-01 in #6109
• [auto-task] Validate the oldest workspace documentation by @orbit-agent-01 in #6112
• doc-duties default selects dir:. and starves on context locks when task-pilot does not narrow it by @orbit-agent-01 in #6111
• Prepare v0.33.0 release by @danieljhkim in #6113

Full Changelog: v0.32.0...v0.33.0