task-plan-v2-orchestrator
Create, refactor, audit, and maintain a multi-agent TASK-PLAN v2 as a canonical Markdown control document with a pre-implementation feature gate, sequential per-task agent orchestration, explicit test/review/rollback fields, wiki sync rules, and a derived HTML dashboard contract. Use when Codex needs to turn loose plans or checklists into structured planning docs, add agent handoff contracts, prepare plans for Codex, Claude Code, Gemini, or Antigravity, or keep Markdown as the source of truth and HTML as a generated dashboard.
How do I install this agent skill?
npx skills add https://github.com/sergekostenchuk/ui-ux-agent-skill-system --skill task-plan-v2-orchestratorIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill is a comprehensive framework for multi-agent task planning and orchestration. It defines strict governance rules to prevent AI hallucinations and ensures transparency by prohibiting hidden mocks or placeholders. No security risks such as malicious code, data exfiltration, or credential exposure were found.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
Task Plan V2 Orchestrator
Use this skill for planning and orchestration artifacts, not for feature implementation itself.
Canonical Model
FEATURE-PREPARATION.mdis the pre-implementation gate.TASK-PLAN.mdis the canonical, editable control document.TASK-DASHBOARD.htmlis derived from Markdown and never the source of truth.
If the user asks for only one file, preserve the same logical separation inside that file.
Workflow
- Classify the request:
- create a new plan
- refactor an existing plan
- audit a plan
- prepare runtime-specific projections
- Read
references/field-layers.md. - If the user mentioned
Codex,Claude Code,Gemini, orAntigravity, also readreferences/runtime-adapters.md. - If the user wants HTML, dashboard, graph, kanban, or observability views, also read
references/dashboard-contract.md. - Start with the pre-implementation layer. If it is incomplete, do not mark implementation tasks as ready.
- Normalize every task into a stable task block with explicit fields instead of freeform prose.
- Model multi-agent execution as sequential by default.
- Add
Execution GovernanceandVerification Policybefore task execution. - Keep missing facts explicit as
TBD,unknown, oropen_question; do not invent them. - Treat critical
TBDfields as blockers forready,in_progress,approved, anddone. - Default to
NO-MOCKSandNO-PLACEHOLDERS. - If a mock or placeholder is temporarily unavoidable, convert it into an explicit alarm with replacement requirements and carry that alarm into every downstream prompt or runtime projection.
Multi-Agent Rules
- A task may have many agents, but only one current
owner_role. agent_sequenceis ordered and sequential unless the user explicitly asks for parallel lanes.agent_contractsmust define:- entry criteria
- expected outputs
- handoff target
- stop conditions
required_approvalsmust name roles or gates, not vague "review".max_review_loopsmust be finite.escalation_rulemust say what happens when review loops are exhausted or a blocker cannot be resolved locally.
Governance Rules
NO-FICTION: do not invent files, commits, checks, approvals, blockers, or artifact paths.NO-MOCKS: do not plan or accept mocks, fake integrations, stubbed results, fake evidence, or pretend backends as a quiet default.NO-PLACEHOLDERS: do not leave placeholders in execution-ready work except as explicit unresolved alarms.INVALID_INPUT: if required input is missing, return the missing fields and do not execute.prompt-first: one task needs one prompt beforein_progress.code-first: implementation tasks cannot close with docs-only progress.strict DONE:donerequires dependencies, approvals, executed verification, evidence, rollback, and sync.verification-first: tests are planned byplanner, reviewed byreviewer, and executed bytester.commands_plannedis not the same ascommands_run.- Placeholder evidence, placeholder tests, and placeholder artifact paths are invalid.
- Unresolved mocks and placeholders must become
alarms, not silent notes. - Every unresolved alarm must state:
- what is mocked or placeholder
- why it exists
- what exact fact, system, credential, artifact, or dependency is missing to replace it
- the replacement target and proposed replacement path
- whether it blocks
ready,in_progress,approved, ordone
- Every unresolved alarm must be copied into:
- the task block
- the task prompt
- the runtime projection
- the handoff summary
What To Read
- Always read
references/field-layers.md. - Read
references/runtime-adapters.mdwhen runtime-specific outputs are needed. - Read
references/dashboard-contract.mdwhen the user wants.htmlor a visual board. - Copy from:
assets/FEATURE-PREPARATION-CHECKLIST.mdassets/TASK-PLAN-v2.template.mdassets/CLAUDE-CODE.tasks-projection.mdwhen a Claude-style condensed tasks view is neededassets/IMPLEMENTATION-PLAN.runtime.mdwhen Gemini or Antigravity needs a projection
Output Rules
- Preserve Markdown as the source of truth.
- Make task blocks stable enough for deterministic parsing.
- Keep a compact register near the top only if it helps navigation or dashboard generation.
- Do not collapse agent-level data into one sentence; keep handoffs explicit.
- When adapting for another runtime, keep field semantics unchanged and only adjust layout or headings.
- If the plan still contains unresolved mocks or placeholders, surface them as top-level warnings and per-task alarms instead of burying them in prose.
- Do not mark a task
ready,in_progress,approved, ordonewhile a critical unresolved mock/placeholder alarm is active.
Runtime Projections
Codex: edit the canonical Markdown directly.Claude Code: keep the canonical Markdown and emit a condensed tasks projection only when useful.GeminiandAntigravity: keep the canonical Markdown and emitimplementation_plan.mdonly as a projection, not a replacement.
Completion Bar
A good TASK-PLAN v2 has:
- an explicit feature-prep gate
- normalized task blocks
- sequential agent contracts
- an
Execution Governancesection - a planned-before-coding
Verification Strategy - test, review, artifact, and rollback fields
- wiki read and write rules
- a dashboard-ready structure
- explicit unresolved-alarm tracking for any temporary mock or placeholder
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/sergekostenchuk/ui-ux-agent-skill-system/task-plan-v2-orchestrator">View task-plan-v2-orchestrator on skillZs</a>