skills.rest
ZK-Theory
ZK-TheoryOFFICIAL
Explore all skills by this author →
GitHub

tda-large-workflow-supervision

Supervises large multi-stage TDL campaigns with capability tracking, dispatch envelopes, and exact-state handbacks.

1|Updated Dec 13, 2025
Search Tags:#git#code review#pull request#handoff#task orchestration#capability tracking#workflow supervision
One-click install
npx skills add https://github.com/ZK-Theory/TDL --skill tda-large-workflow-supervision-zk-theory
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: tda-large-workflow-supervision
Source: https://github.com/ZK-Theory/TDL/tree/main/.agents/skills/tda-large-workflow-supervision
Command: npx skills add https://github.com/ZK-Theory/TDL --skill tda-large-workflow-supervision-zk-theory

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Large, multi-stage engineering campaigns lose state when one conversation becomes the state store, leading to premature completion claims, lost context after compaction, and unreviewed merges. This Skill provides a standalone supervision protocol that keeps the named end-to-end capability as the unit of completion across rotations, PRs, and reviews. ## Core Features & Use Cases - Capability Campaign Contract: Tracks capability states (NOT RUNNABLE through INTEGRATED) and forbids promoting a locally complete slice to campaign completion. - Dispatch Envelopes: Records exact base, subject, writable root, bounded contribution, context mode, and validation boundaries for every implementer or reviewer dispatch. - Certify/Deliver Phases: Reconstructs a campaign once into an identity-based packet, then delivers from fresh tasks with delta verification instead of full replay. - PR Packaging Rules: Enforces the CodeRabbit 100-file cap, semantic split boundaries, and commit-bound review verification before merge. - Use Case: Supervising a multi-PR feature campaign where an external reviewer is pending: persist a PR-ready handoff, close the task as OWNER-BLOCKED, and resume from a fresh integration-closer task after the review completes. ## Quick Start Ask the AI to supervise a standalone TDL campaign by declaring the supervision phase, naming the end-to-end capability, and recording its state and next production action.

Frequently Asked Questions about tda-large-workflow-supervision

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I supervise a large multi-stage workflow without losing state?▼

Declare a standalone workflow with a certify or deliver phase, set one named end-to-end capability with its state and next production action, and rotate to fresh tasks only after actual compaction. Record exact-state handoff packets at each rotation so no single conversation becomes the state store.

When should I use standalone supervision instead of APM lifecycle skills?▼

Use standalone supervision when the campaign is not APM-managed and has no .apm campaign state or Memory Bank. Generic role names like Manager do not establish APM ownership, so numbered APM skills and APM guides must be rejected in standalone mode.

How do I handle CodeRabbit review limits on large pull requests?▼

Count changed paths with git diff --name-only before opening a PR and keep it at 90 files or fewer where practical, since CodeRabbit caps at 100. Split larger work on semantic dependency boundaries, keep one campaign status across the PRs, and verify the reviewed commit matches the current PR head before merging.

What context mode should implementer and reviewer tasks use?▼

Independent reviewers receive no parent history, while implementers use the least costly mode that carries required evidence: a fresh task for new subjects or a bounded continuation when replaying evidence costs more. Never use full-history inheritance after actual compaction.

When is a capability campaign actually complete?▼

A campaign is complete only at INTEGRATED status, after a real end-to-end production path has been exercised on the assembled result. Plans, synthetic evidence, intermediate PRs, and locally complete slices never establish completion, and a second final acceptance remediation triggers rescope instead.

What should I do while waiting for an external review to finish?▼

Treat the owner-controlled review wait as a task boundary: persist a durable PR-ready handoff with the exact accepted subject, tests, and unblock condition, then close the task as OWNER-BLOCKED. Resume later from a fresh lightweight integration-closer task after the owner reports completion.

Related Skills

Explore More Software Engineering Skills→
cenjiecenjie

i-shape

Plan UX and UI through a structured discovery interview and design brief synthesis.

Community
Intermediate
synthaicodesynthaicode
4

goal_mode

Preserves restart-ready task state and resumes Codex work after quota recovery.

Community
Intermediate
DeiblerDeibler
8

prototyping

Guides selection of prototype fidelity and type to validate design assumptions before implementation.

Community
Intermediate
stephendorstephendor
1

apm-5-check-reports

Check APM Manager Task Reports across Report Buses for pending items.

Community
Intermediate