agentsclimarketplace

Spec phase orchestration

Skill wei18/apple-dev-skills/collaboration-skills/skills/spec-phase-orchestration

Reusable Claude Code skills for AI-agent-driven Swift / Apple-platform development — composable via git submodule; aggregates other specialist skill repos

Install
npx -y skills add wei18/apple-dev-skills --skill spec-phase-orchestration

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

The pre-implementation document pipeline — 5 files + `meetings/` directory (README.md + docs/design.md + docs/foundations.md + docs/plan.md + docs/methodology.md + meetings/), section-by-section approval (§What before §How), prerequisite checklist with Unconfirmed / Resolved gates, "no implementation code before design.md and plan.md approved" rule. Invoke when starting a new project that needs spec-first development, deciding doc structure, or when asked "which documents go in the spec phase".

SKILL.md

5.1 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it

Spec Phase Orchestration

When to invoke

  • Starting a new project and deciding which documents the spec phase produces.
  • Multiple documents / sections are in flight and you need to decide the order of progress.
  • User asks "design or foundations first", "can we proceed with prerequisites unresolved".

Default decisions

5 files + meetings/ directory

<project>-spec/
├── README.md                  # Project entry point
├── docs/
│   ├── foundations.md         # Engineering substrate (language version, modules, CI, L10n, secrets)
│   ├── design.md              # Product spec §What + technical design §How (unified)
│   ├── plan.md                # TDD-ordered checklist
│   └── methodology.md         # Claude agent application practices (living doc)
└── meetings/
    └── {YYYY-MM-DD}_{topic}.md

Don't split further:

  • No separate spec.md / rfc.md / tasks.md (folded into design.md / plan.md)
  • No adr/ directory; major decisions land directly in design.md or foundations.md (open one only if real need arises)

Progression order

foundations.md §1..§N (one section at a time) →
design.md §What (product spec) →
design.md §How (technical design) →
plan.md (TDD-ordered) →
implementation phase begins

Rule: don't write §How before design.md §What passes; don't write implementation code before plan.md passes.

Section-by-section approval

  • The Leader advances one section at a time, waiting for user / review approval before moving on.
  • Multiple sub-agent rounds may happen within a single section (see subagent-review-cycles).
  • Section order is non-skippable: §What before §How, §1 before §2.

Prerequisite checklist (Unconfirmed / Resolved gates)

Any proposal depending on external tools / APIs / third-party packages must include a prerequisite checklist:

**Prerequisites**:
- [ ] Behaviour of Apple `XYZAPI.foo()` is consistent across iOS 26 and 25 — **Unconfirmed** (verified in plan.md step N)
- [x] ~~Xcode Cloud hook naming~~ — **Resolved** (Code Review round N, date): adopt `ci_post_clone.sh`
  • Unconfirmed items block Leader approval — the section stays DRAFT until each item is Resolved.
  • Resolved items are checked off with their decision basis + date.

No implementation code before design + plan approved

  • Before design.md §What + §How pass and plan.md is written and approved, no implementation Swift code is written.
  • Exception: a pure exploratory spike — must be logged under meetings/ as a spike; its output doesn't land on main.

Backlog sections

Every living doc carries a §Backlog. Stray ideas are routed by topic: "product → design", "tooling → foundations", "implementation step → plan", "collaboration → methodology", "unclassifiable → meeting log" (see backlog-routing-by-topic).

Rationale

  • 5 docs + meetings dir is the "enough to record, not redundant" sweet spot for solo / small teams; 8 files is over-structured, 4 lacks room for foundations / methodology.
  • Section-by-section avoids the high cost of "send the whole doc back".
  • Prerequisite gates force "assumptions" to become "items to verify"; CLI trial-and-error is not the default answer.
  • Spec-first: "no spec, no code" is a fundamental premise of AI agent collaboration (otherwise prompts are vague and output diverges).

Deviation considerations

  • Tiny utility / 1-day project: keep only README.md + meeting log; inline the design in the README.
  • Adding a feature to an existing codebase: skip foundations.md; lean on the existing conventions.
  • Multi-person collaboration: plan.md may split into plan/<feature>.md, but keep a master plan listing the dependencies.
  • Full outsourcing / contractor: use an RFC + tasks split instead; this skill is a solo / small-team default.

Verification checklist

  • 5 docs + meetings dir structure in place (README + 4 docs + meetings dir).
  • design.md contains §What and §How; §How is only filled after §What passes.
  • foundations.md sections carry Status (DRAFT / FINAL / OBSOLETE) + Open items (if any).
  • plan.md steps are TDD-ordered (test step precedes implementation step).
  • Every prerequisite item is explicitly marked Unconfirmed / Resolved.

Related skills

  • backlog-routing-by-topic: rules for placing stray ideas.
  • subagent-review-cycles: round structure inside spec sections.
  • leader-developer-handoff-contract: dispatch contract when handing sections to a sub-agent.
  • session-to-meeting-log: persisting each spec session.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.