ワンクリックで
kyberbot-factory-console
Operator console for the kyberbot factory — create, manage, and advance jobs through the pipeline.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Operator console for the kyberbot factory — create, manage, and advance jobs through the pipeline.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Operator console for the kyberbot factory — create, manage, and advance jobs through the pipeline.
Read-only queries across kyberbot factory jobs — status, search, and reporting.
Persist important information from this conversation to long-term memory. Use proactively whenever the user mentions a person, project, company, decision, meeting, deadline, preference, or any fact that future sessions should know about. Also use when the user says remember this, store this, note this, keep track of, or don't forget.
Look up what the agent knows about a person, project, company, place, or topic from the entity graph and timeline. Use proactively whenever the user mentions someone by name, references a project or company, asks about past interactions, or says who is, what do we know about, tell me about, or recall.
Back up all agent data to GitHub — SQLite databases, Claude Code memory, identity, skills, brain notes. Use when the user says backup, back up, save everything, push to github, or snapshot. Also used by heartbeat for automated backups.
Save long-form knowledge, research findings, architecture decisions, reference material, or structured notes to the brain/ directory for persistent retrieval. Use when the user shares detailed information that doesn't fit a single memory, discusses architecture or design decisions, provides reference material, shares meeting notes or research, or says save this to the brain, write this down, document this, or take notes on.
| name | kyberbot-factory-console |
| description | Operator console for the kyberbot factory — create, manage, and advance jobs through the pipeline. |
Operator console for the kyberbot factory. Surfaces the attention queue, handles job creation, and moves jobs through the pipeline using the universal operator action pattern.
Scan kyberbot-factory/jobs/ for jobs in operator-owned states:
| State | Meaning |
|---|---|
draft-input | Draft gate flagged missing detail — operator must fill in PURPOSE / REQUIREMENTS / type before pipeline can start |
research-input | Research agent has questions |
plan-input | Planning agent has questions |
test-input | Testing agent failed the suite, couldn't author a needed test, or hit ambiguity — operator triages |
Newly-filed jobs (state drafted) are not in the attention queue — they are picked up by the dispatcher and routed by the draft-gate agent. If the gate finds the job sufficient, it advances straight to research; if not, the job lands in draft-input for operator clarification.
Present via AskUserQuestion:
[{STATE}] {id} {title} (e.g., [PLAN-INPUT] KF-002 Add WebSocket support).If no items require attention, show only "Create new job" and "Exit".
The operator selects an item (entering the action pattern) or creates a new job.
When the operator selects a job, read the delamain transitions where from = current state. Group by class and present via AskUserQuestion:
| Action | Description | When shown |
|---|---|---|
| Review | Read the job file and display it inline | Always |
| Respond | Do the state-specific work, then choose a transition | If any advance or rework transitions exist |
| Shelve | Shelve the job | Always |
| Cancel | Cancel the job | Always |
| Exit | Return to attention queue | Always |
If a state has no advance or rework transitions, omit the Respond action.
Read the full job file with the Read tool and display its content for the operator. After display, return to the action menu for this job.
Respond is a two-phase action: do the work, then choose direction.
What "do the work" means depends on the state:
| State | Phase 1 |
|---|---|
draft-input | Read the latest draft-gate: ACTIVITY_LOG entry to see which sufficiency checks failed. Walk the operator through filling in the missing pieces (PURPOSE, REQUIREMENTS, type) directly in the job file. Then advance back through drafted so the gate re-evaluates. |
research-input | Answer research agent questions (read RESEARCH_QUESTIONS, present to operator, collect answers, append) |
plan-input | Answer planner agent questions (read PLAN_QUESTIONS, present to operator, collect answers, append) |
test-input | Read the latest dated entry in TESTS plus the most recent testing: ACTIVITY_LOG line. Present the failure / ambiguity narrative to the operator. The triage path is: rework → dev (failures the dev agent should fix) or advance → done (operator override — known flake, environment-only failure, accepted gap). There is no rework-to-testing transition; if the operator wants the testing agent to retry, they pick rework → dev with a no-op note for the dev agent. |
Phase 1 is guided, not delegated. Walk the operator through each step — present one question at a time, collect the response, run verification where possible, then move to the next. The agent is an active partner, not a questionnaire.
After the work is done, read the delamain transitions from the current state and present the legal options:
| Transition class | Presented as |
|---|---|
| advance | Advance to {state} — one option per target |
| rework | Rework to {state} — one option per target |
If a class has multiple targets, show one option per target. If a class has zero targets, omit it.
updated to today's date## ACTIVITY_LOG with transition contextgit add -A && git commit -m "kyberbot-factory: {id} {from-state} → {to-state}"
After the commit, return to step 1 (re-scan the attention queue).
Collect from the operator:
feature, enhancement, defect, hotfix, security, or choreSet defaults:
id: scan kyberbot-factory/jobs/ for existing files, determine the next highest integer, format as KF-{NNN}status: draftedcreated: todayupdated: todayCreate the job file at kyberbot-factory/jobs/{id}.md using the job entity shape. Append to ACTIVITY_LOG: - {today}: Created.
Run:
git add -A && git commit -m "kyberbot-factory: {id} created"
Then return to step 1 (attention queue). The dispatcher will pick the new job up automatically — draft-gate runs first and either advances the job to research (if PURPOSE / REQUIREMENTS / type all check out) or routes it to draft-input for operator follow-up.
kyberbot-factory-inspect).