cratis-screenplay-delivery-planning
Plan and estimate the delivery of an accepted Screenplay model by slices: count slices by type (StateChange, StateView, Automation, Translate), derive the slice dependency table from the model, split work in parallel along event contracts, track slice groups as GitHub issues with a closed status vocabulary, and turn the team's own completed-slice history into ranged estimates, capacity math and retrospective metrics. Use when asked how long a modeled scope will take, how to split it across people or agents, what to build first, or how to track and re-plan delivery. Not for: deciding what to model (use `cratis-screenplay-event-modeling`), rendering or gap-filling a slice (use `cratis-screenplay-render-and-gap-fill`), or reviewing the model (use `cratis-screenplay-model-review`).
- License 许可
- MIT
- 所属仓库
- Cratis/Prologue
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- cratis-screenplay-delivery-planning
- license
- MIT
- description
- Plan and estimate the delivery of an accepted Screenplay model by slices: count slices by type (StateChange, StateView, Automation, Translate), derive the slice dependency table from the model, split work in parallel along event contracts, track slice groups as GitHub issues with a closed status vocabulary, and turn the team's own completed-slice history into ranged estimates, capacity math and retrospective metrics. Use when asked how long a modeled scope will take, how to split it across people or agents, what to build first, or how to track and re-plan delivery. Not for: deciding what to model (use `cratis-screenplay-event-modeling`), rendering or gap-filling a slice (use `cratis-screenplay-render-and-gap-fill`), or reviewing the model (use `cratis-screenplay-model-review`).
02 资源文件 5
- 许可证LICENSE597 B
- 参考文档references/parallel-delivery.md4.2 KB
- 参考文档references/provenance.md3.6 KB
- 参考文档references/slice-counting.md3.0 KB
- 参考文档references/velocity-and-capacity.md4.0 KB
03 SKILL.md 正文
<!-- cratis-ai-managed: skills/cratis-screenplay-delivery-planning/SKILL.md -->
Delivery planning
Purpose
Turn an accepted .play model into a delivery plan: how many slices of which type, in which
order, who can work in parallel, what each group costs, and when it is done. The model already
cuts the work into slices and names the events between them, so the plan counts and orders
those slices instead of guessing at features.
Planning does not replace modeling, review or the render and gap-fill gates. A plan is a working note, not a contract: when the model changes, recount; when a slice is not accepted, it is not planned work yet.
Ground rules
- Count slices, not features. The unit is one slice of one type
(
StateChange,StateView,Automation,Translate). Method and weights:references/slice-counting.md. - Estimates are ranges from the team's own history. State the basis (which completed slices, how many, how long). With no history the estimate is a first estimate without history: say so, give a wide range, and replace it as soon as slices complete. Never quote an accuracy figure or a duration you cannot trace to a count and a rate.
- Events are the contract. Slice B depends on slice A because B needs an event or view A
produces. Anyone can build B against the events A will produce, with synthetic
givenevents, before A is finished. Dependencies are read from the model, not remembered. - Render is not free of effort, and gap-fill is not free of risk. Stage renders
StateChangeandStateView;AutomationandTranslateare hand-written against the model as contract. Check admission per slice withcratis-screenplay-render-and-gap-fill(tool versions:cratis-screenplay-toolchainSKILL.md) before pricing a slice as rendered. - Plans live in
.ai-work/; tracking lives in GitHub issues where the repository works that way. Never put a plan in the model, in.cratis/, or in a documentation folder.
Procedure
- Confirm the scope is accepted. Only slices of a committed model under the root count as planned delivery. Drafts and open questions are listed separately as unplanned. If the model is still moving, plan only the stable modules and say what is excluded.
- Inventory the slices. Use the Screenplay MCP:
describe-applicationfor counts and paged children,dependency-graphat slice level for edges, cycles and an advisory order,dependenciesandfind-referencesper slice for the contracts it consumes and produces and for theinvokes, action and form edges the graph does not order (readsmaps todecidesFrom, which it does order, but it stays a coordination point). Treatcomplete: falseor an ambiguous binding as unknown, never as no dependency (cratis-screenplay-model-authoring). Without the MCP, read the.playfiles. - Classify each slice by type and by realization: rendered (Stage admits it) or hand-written (gap-fill, adapter, external call). Record the reason for each hand-written one.
- Build the dependency table (
references/parallel-delivery.md): per slice, the events it consumes, who produces each, and the slices it waits on. A cycle means a feedback loop: return it to slice design (usually an Automation slice); never schedule around it. - Group into waves. Slices with no unmet producers form wave 1; each next wave needs only earlier waves. Within a wave, slices are independent and can run in parallel. Group slices that share one view or one stream into one slice group (one issue).
- Weigh and estimate (
references/velocity-and-capacity.md): apply the team's weights per type and realization, divide by measured throughput, and report a range with its basis. Add separate lines for what is not slice work: modeling and review gates, integration with outside systems, environment and deployment, specifications that need new infrastructure. - Track. One GitHub issue per slice group, in the repository's own issue practice
(assign the issue, keep an in-progress status in the issue body, check for an existing claim
before starting): the status block lists the slices, their
type, the wave, the blockers and each slice's planning status below. The issue's own state
follows the repository's hygiene (in progress, in review when the PR opens, paused, blocked,
done only when delivered). Keep the working plan in
.ai-work/<task>/delivery-plan.md. - Re-plan after each completed group. Update throughput, recount if the model changed,
restate the range. Record the retrospective metrics from
references/velocity-and-capacity.md.
Status vocabulary
Per-slice planning status, a closed set recorded inside the plan and the issue's status block. It is not the issue's state: the enclosing issue keeps the repository's own states (for example in progress, in review, paused, blocked, done), and is done only when its slices are delivered (merged and released where the repository releases).
| Status | Meaning |
|---|---|
| todo | accepted in the model, not started |
| in progress | one owner is working on it |
| blocked | waiting on a named producer, decision or external party; say which |
| done | built, its specifications pass, and the slice conforms to the model |
Before claiming a slice, check its current status. If it is already in progress, someone else
holds it: pick another slice, do not take it over and do not report an error. Move a status only
to a different value; a repeated update is a signal that another worker got there first.
Check for an existing claim (assignee and in-progress status) before starting; where the harness
provides a claim tool such as issue_work, its claim step is that guard.
What a plan states
- Scope: the modules and slices counted, and what was excluded and why.
- Slice counts by type, split into rendered and hand-written.
- The dependency table and the waves.
- The estimate as a range with its basis, or "first estimate without history".
- Capacity assumptions: who or what works in parallel, and the constraints on it.
- Risks that move the range: hand-written slices, outside systems, open questions, a model likely to change.
Limits
- A rate from one project or one person does not transfer to another team or another kind of slice; say when the sample is small or mixed.
- Counting hides unusually large slices. A slice that needed a design discussion is two slices or a modeling question; recount rather than weigh it as "big".
- Parallel work only pays when slices are truly independent in the dependency table and integration of the shared events is owned. Shared writes to one file, one stream or one external system serialize the work.
References
references/slice-counting.md- types, realization, weights, what to count and exclude. Read when counting or classifying slices.references/velocity-and-capacity.md- measuring throughput, ranges, capacity math, retrospective metrics. Read when producing or updating an estimate.references/parallel-delivery.md- dependency table, waves, event contracts, issue grouping. Read when splitting work across people or agents.references/provenance.md- sources and what was not adopted.
Related skills
cratis-screenplay-slice-design (slice dependencies), cratis-screenplay-model-authoring
(MCP navigation), cratis-screenplay-render-and-gap-fill (what renders, gap-fill handoff),
cratis-screenplay-modeling-lifecycle (phases and gates).
04 同仓库更多 Skills
vectorization
Designs, implements, optimizes, and reviews SIMD code in .NET: vectorizing scalar loops with TensorPrimitives, Vector64, Vector128, Vector256, Vector512, or hardware intrinsics, and reviewing existing SIMD code, including the generic Vector type, for contract equivalence, tail handling, memory safety, portability, fallbacks, and measured performance. Use when a .NET loop or SIMD routine is being written or reviewed. Not for performance work unrelated to SIMD.
target-authoring
Canonical patterns for writing custom MSBuild targets: DependsOnTargets versus BeforeTargets and AfterTargets, the Build-to-CoreBuild three-level pattern, and the XxxDependsOn chain-extension pattern. Use when diagnosing or fixing custom-target anti-patterns: Directory.Build.targets redefining SDK targets, targets that replace CompileDependsOn instead of extending it, Outputs versus Returns misuse that returns stale results, missing Inputs/Outputs causing needless rebuilds, or missing FileWrites registration. Not for incremental build tuning or general anti-pattern review (use the msbuild skill), or non-MSBuild build systems.
run-tests
Chooses and runs the repository-compatible .NET test command: classic, VSTest, bridged MTP, or native MTP syntax for MSTest, xUnit, NUnit, and TUnit. Use before running .NET tests or answering with a test command or flags: run the tests, one test, class, category, trait or target framework, combined filters, --filter-query, --no-build, --diag, TRX, coverage collection, crash or hang dumps, filter errors, or unrecognized options. Not for platform identification alone (use platform-detection), writing or debugging test code, interpreting a coverage report (use coverage-analysis), CI investigation, migration, or a persistent hot-reload loop.
profiling
Uses the free official .NET diagnostics CLI tools for profiling and runtime investigation in a repository: CPU, GC pressure, allocation growth, exception storms, lock contention, and slow code. Use when a .NET application needs performance or runtime profiling. Not for replacing realistic performance or load tests with ad-hoc tracing, production heap collection without a pause-risk plan, crash dumps (use dump-collect), or benchmark design (use microbenchmarking).
platform-detection
Identifies a .NET project's test platform, framework, command mode, and SDK-style versus classic project system, resolving global.json, project, packages.config, Directory.Build.props, and Directory.Packages.props precedence for MSTest, xUnit, NUnit, and TUnit. Use when asking "which test platform or framework?", "VSTest or MTP?", or "what runner does this project use?", including bridge settings, UseVSTest opt-outs, and conflicting VSTest/MTP configuration. Not for running or filtering tests, commands, flags, TRX, dumps, or test-command errors (use run-tests), or for hot-reload or migration requests.
orleans
Designs Microsoft Orleans systems from each primitive's purpose and failure model: grains, digital twins, state versus databases, transactions, messaging, streams, timers, reminders, Durable Jobs, stateless workers, grain services, startup, and hosting. Use when a task needs an Orleans design decision: inspect version and topology, choose the primitive, implement, and validate. Not for other actor stacks, batch processing, relational-only CRUD, or advice without an Orleans decision.