Mission Control

24h Ops

criticalAll agents stopped — open Cursor/Codex and say "继续"0/h

0/2 reporting2 stale0 findings1 waiting

MC Next Actiondrive

Drive mission-control: MC product/control-plane QA

Track pickup in COLLAB and route unclear ownership immediately.

Since Last Visit0 commits · 10 decisions

decision Typed actor collaboration templates defined — jiansen + codex

decision Elegant Actor Workflow v1 implementation slice def — jiansen + codex

decision R-008 upgrade control plane phase 2 landed — codex

decision R-008 registration/enrollment API contracts landed — codex

decision R-008 event ingestion and machine freshness landed — codex

decision R-008 pairing token contract landed — codex

+5 more in timeline below

1 decision waiting for you

action required

R-006mission-controlblocks: Vibeful GitHub Actions budget / private repo wrapper CI red

GitHub annotation on Vibeful wrapper jobs 26017933766 and 26015833613: The job was not started because an Actions budget is preventing further use. The real release builds in Jiansen/vibeful-releases are separate: v0.2.46 real run 26015835891 succeeded and v0...

Project summary: green readiness · 0 active findings · 0 commits/24h

2 stale runtime report(s)Cursor resume:cd ~/Coding/mission-control && cursor agent --continueCodex reports on next automation heartbeat.or copy all Cursor commands: collab kickall

Target a role (builder/verifier) instead of a specific tool · Full Command → · / focus to

Actor Fleetcodex@mission-controlCodexstale report1d agocursor@mission-controlCursorstale report1d ago← 继续0 reporting · 2 stale · 0 inferred

Fleet detail

Actors	Kind	Runtime	Active Role	Proof	Queue Policy
codex@mission-control	agent	Codex	verifier	not configured	next-heartbeat-confirm
cursor@mission-control	agent	Cursor	builder	not configured	immediate-init

Patrol Decision Detail: Drive mission-control: MC product/control-plane QAcodex · drive

Codex patrol is conservative on empty queue: if no self-block is present, wait for next heartbeat/re-read before initiating a new high-ROI task, unless board drift/security/authorized verification is obvious.

Track pickup in COLLAB and route unclear ownership immediately.

Project	Readiness	Owner	Gates	Findings	24h	Last

mission-control

ready

unknown @ unknown · 1d ago

0active

0verified

024h

0dirty

Evidence (0 findings, no release)

🟢 多项目 COLLAB.md 聚合（vibeful + beefit） Command Center summary（needs-you / agent proof / project pulse / recent change / MC next action；legacy `/morning` route remains an alias only） Decision Queue + inline reply（写入 COLLA...

Commits (24h)

All registered repos combined.

Active findings

Still need drive or verification.

Agent reporting

0/2

0 active, 2 stale, 0 inferred

Projects

mission-control: 0

Last human input

2026-05-20

Most recent jiansen decision in log

Recent Activity (last 6h) · 0 commits

No recent commits.

Control Plane Details (Runtime · VCP · Reconciliation · Dependencies · Stale Ownership)

Agent Runtime

0 active · 0 idle · 2 stale

Reports: 0 explicit · 0 fallback · 2 stale

Adoption: 0 need reports · 0 already routed · 0 unrouted

Command source: 0 need shared collab · 0 routed · 0 unrouted

0 explicit reports

Runtime rows prefer explicit agent reports from docs/agent-runtime.json + child .collab/runtime/reports.jsonl; repository activity and protocol files are fallback evidence only.⚠ 1 stale rule file(s) detected (>7d unchanged)

codex@mission-controlstale

1d ago · patrol:on · ask:disabled

agent-runtime.json: collab report-runtime

patrol:onreport:stalerules:freshcollab:sharedempty:next-heartbeat-confirmproto:shared COLLAB patrol protocol

task done: not reported

protocol review: not reported

self-block handled: not reported

Refresh stale agent-runtime report (>30m old)

cursor@mission-controlstale

1d ago · patrol:on · ask:disabled

agent-runtime.json: collab report-runtime

patrol:onreport:stalerules:freshcollab:sharedempty:immediate-initproto:shared COLLAB patrol protocol

task done: 1d ago

protocol review: 1d ago

self-block handled: 1d ago

Refresh stale agent-runtime report (>30m old)

Verification Control Plane

0 pending · 0 ready · 0 stale · 0 conflicts

journey gates

Aggregates verification pipeline health by product journey. Shows which claims are stale, where evidence is missing, board conflicts, and untestable items blocked by upstream dependencies.

Browser Task New Taskuntestable

A user can create and start an installed Browser Task from the desktop app without security prompts or empty-task artifacts.

vibeful.browser-task.new-task·vibeful agents

✗Installed app opens the New Task modal from the user-facing entry points.

✗Goal is required and empty/title-only creation is rejected.

✗Create and start launches Chrome for Testing without Keychain or profile prompts.

✗Task reaches a terminal status with visible output.

✗Browser sessions and task-profile processes clean up after completion.

→ Register vibeful and ensure COLLAB.md exists

⛔ Project vibeful not found in registered projects

Browser Task Parallel Isolationuntestable

A user can run three App-created browser tasks concurrently without sidepanel, tool, or profile cross-contamination.

vibeful.browser-task.parallel-isolation·vibeful agents

✗Three tasks are created from the installed App UI, not only local dev tooling.

✗Each task launches a distinct browser window/profile and binds to the correct sidepanel task.

✗Agent tool calls for one task do not affect the other tasks' browser state.

✗All three tasks complete with outputs.

✗Sessions return to empty and no orphan Chrome-for-Testing task-profile process remains.

→ Register vibeful and ensure COLLAB.md exists

⛔ Project vibeful not found in registered projects

Browser Task Report Qualityuntestable

A realistic long research task produces a usable sourced report without premature wrap-up language.

vibeful.browser-task.report-quality·vibeful agents

✗Task is realistic enough to require multiple browser actions and synthesis.

✗Report includes useful structure, source references, and actionable findings.

✗No budget or wrap-up copy degrades output quality before the task is actually complete.

✗Final task status, plan progress, and output state agree.

→ Register vibeful and ensure COLLAB.md exists

⛔ Project vibeful not found in registered projects

Browser Task Session Cleanupuntestable

Task completion, deletion, resume, and browser close leave the runtime reusable for the next task.

vibeful.browser-task.session-cleanup·vibeful agents

✗Completed tasks do not leave live browser sessions.

✗Deleted active tasks terminate MCP sessions and browser processes.

✗Resume does not attach to stale locked profiles.

✗Post-completion browser state does not show crash pages caused by forced process teardown.

→ Register vibeful and ensure COLLAB.md exists

⛔ Project vibeful not found in registered projects

Desktop Update UXuntestable

Installed users see accurate update state and actionable updater phases.

vibeful.desktop.update-ux·vibeful agents

✗Installed version, latest release, and updater latest.json do not produce stale downgrade banners.

✗Available update points to an installable non-draft release.

✗Updater download/install phases are visible.

✗Stalled download becomes an actionable retry/error state.

→ Register vibeful and ensure COLLAB.md exists

⛔ Project vibeful not found in registered projects

Agentic Audit Verifier Ledgeruntestable

Operators can see why LLM verifier claims are trusted or rejected for a task.

beefit.agentic-audit.verifier-ledger·beefit agents

✗Task-level verifier summary is visible and fresh.

✗Mapping-review exposes per-field llm_verdict and llm_reason where claimed.

✗Verifier evidence is tied to task id and run timestamp.

✗Timeout or missing verifier output is represented as incomplete, not trusted.

→ Register beefit and ensure COLLAB.md exists

⛔ Project beefit not found in registered projects

Agentic Audit Action Safetyuntestable

AI can explain and propose audit actions without silently mutating production state.

beefit.agentic-audit.action-safety·beefit agents

✗Fresh chat receives current anomaly context.

✗AI explains the correct unresolved anomalies.

✗AI actions are proposal-first and role-safe.

✗A user-accepted safe action persists the expected cell/status mutation.

✗Rejected or unauthorized actions do not mutate task state.

→ Register beefit and ensure COLLAB.md exists

⛔ Project beefit not found in registered projects

Audit Progress Contractuntestable

Progress, current stage, and unresolved work tell ordinary operators what to do next.

beefit.audit.progress-contract·beefit agents

✗Ready tasks with unresolved work do not show 100% complete.

✗Submitted/review/approved states use distinct progress and copy.

✗Current stage and next action match unresolved work.

✗Task page and progress API agree.

→ Register beefit and ensure COLLAB.md exists

⛔ Project beefit not found in registered projects

Mapping Prescan Semanticsuntestable

Prescan and mapping review pick the correct business columns in repeated-section Excel sheets.

beefit.mapping.prescan-semantics·beefit agents

✗Repeated sections such as production-material blocks are disambiguated by headers and samples.

✗LLM-provided column indexes are validated against headers before use.

✗Core aluminum costing fields map to the business columns, not adjacent context columns.

✗A second sample beyond the original failing task passes the same semantic checks.

→ Register beefit and ensure COLLAB.md exists

⛔ Project beefit not found in registered projects

Mapping ColumnRef Contractuntestable

Engine, API, LLM prompts/actions, and UI use one user-facing Excel-letter plus header contract beyond Z.

beefit.mapping.column-ref·beefit agents

✗API returns Excel letters for mapped fields, including AA/AB beyond Z.

✗UI displays Excel letters plus headers consistently.

✗LLM action schema accepts the same ColumnRef format.

✗Conflicting numeric/letter references are rejected before mutation.

✗Valid AB remap works on a disposable or restorable task.

→ Register beefit and ensure COLLAB.md exists

⛔ Project beefit not found in registered projects

All verification pipelines clear. No stale claims or conflicts detected.

Board Reconciliation

0 critical · 0 drive · 0 watch

control-plane repair queue

Converts VCP conflicts, stale dependencies, ownership drift, and runtime-report adoption into one repair queue. These are MC infrastructure actions: fix the board, route the owner, or wait on an already-open Request.

No reconciliation work detected. Control-plane boards are internally consistent.

Verification Readiness

0 ready · 0 blocked · 0 waiting

COLLAB verify

Converts `fixed_pending_verify`, `pending_codex_verify`, `claim`, and `verify` rows into an action list for Codex. Waiting rows need another owner to claim evidence first.

No verification-ready rows detected. Continue monitoring claims and release handoffs.

Stale Ownership

0 drive · 0 watch

COLLAB route hygiene

Finds active rows without an accountable owner, actionable rows without a concrete next action, and closed rows that still look routed. Use the inline route form for single-row correction.

No ownership drift detected in active or recently closed finding rows.

Dependency Map

0 blockers · 0 edges · 0 blocked findings

COLLAB blocked_by

Derived from each project's `blocked_by` / `blocking` fields. Use this to maintain dependency hygiene: unresolved upstreams, verified upstreams still referenced as blockers, and high-impact roots should be fixed or rerouted first.

No dependency edges detected.

Mission Control Self

unknown @ unknown

green

commits 24h

queues

requests

dirty

Current MC work

Product/access decisions only

No stale MC design blocker. User has corrected direction: 24h Operations Console and continuous project management, with Command Center as action surface.

Project Registry

3 registered · 0 discovered

docs/projects.json

beefit

registered

/Users/jiansenhe/Coding/beefit/docs/COLLAB.md

docs/projects.json

Registered but COLLAB.md is missing or unreadable.

mission-control

registered

/var/task/docs/COLLAB.md

docs/projects.json

Loaded into dashboard metrics and patrol decisions.

vibeful

registered

/Users/jiansenhe/Coding/vibeful/docs/COLLAB.md

docs/projects.json

Registered but COLLAB.md is missing or unreadable.

Product Direction

draft-ready

Open draft

Build Mission Control as an agent ops console first, with a triage-board layer for handoffs. It should feel closer to Vercel project operations plus Linear/Sentry triage than to a generic analytics dashboard.

First viewport should expose the current patrol decision, Mission Control self-health, project registry/discovery, and cross-project risk. Project cards and queues sit below that, because the primary user question is "what needs driving now?" not "how many cards exist?"

/var/task/docs/product-design.md

Work Queues

mission-control · jiansen

Product/access decisions only

No stale MC design blocker. User has corrected direction: 24h Operations Console and continuous project management, with Command Center as action surface.

mission-control · codex

MC product/control-plane QA

mission-control · cursor

R-009 done, patrol continues

Cursor patrol is aggressive on empty queue: after task completion and re-read, handle self-blocks immediately; if none, start a highest-ROI initiative task in the same long-running request/session. R-009 shipped; next: ...

mission-control · child projects

Vibeful/Beefit self-iteration

Subproject findings should be owned by their own agents. MC may sample evidence and write routing/Requests, but should not make subproject E2E execution its primary work unless needed to validate MC tooling itself.

Agent-to-Agent Requests

All agent coordination resolved — no pending requests.