cli-forge-intake

Installation
SKILL.md

cli-forge Intake

Use this stage when the current request needs routing or when you want to resume the cli-forge workflow from the earliest safe phase.

Stage Goal

Finish this stage with:

  • the work classified as description, scaffold, extend, validate, or publish
  • publish-intent requests narrowed to the correct child skill: cli-forge-publish for repo-native release work or cli-forge-publish-npm for optional npm publication
  • the required user inputs identified and, when available, assembled
  • the shared constraints loaded from the planning brief and the relevant instruction document
  • an explicit handoff to the next stage skill

Canonical References

When To Use This Stage

  • The user asks for cli-forge help in a high-level way and the correct phase is not obvious yet.
  • The user wants to create a new project but has not clearly supplied a valid skill name.
  • The user wants to create a new project and the generated skill's description contract has not been approved yet.
  • The user wants to add stream or repl, but the project path or feature is still unclear.
  • The user wants to update an existing generated skill's purpose, positioning, or other user-facing contract.
  • The user wants validation, but the target project path still needs to be resolved.
  • The user wants to publish, do a release dry run, rehearse the destination mirror, check destination configuration, adopt the release automation asset pack into a target CLI skill project, or publish the shipped CLI command to npm.

Required Inputs By Outcome

  • description: the classified intent, generated skill scope, and the next stage that will consume the approved description contract
  • scaffold: skill_name, plus optional author, version, and rust_edition
  • extend: project_path and feature where feature is stream or repl
  • validate: project_path
  • publish: release mode (report_only, dry_run, rehearsal, or live_release), requested publication channel (repo_native, npm, or both), explicit target-repository context, current validation status, and any required destination configuration, npm package-set context, or credential context

Workflow

  1. Read the user request and classify the intent as description, scaffold, extend, validate, or publish.
  2. Load ./planning-brief.md before choosing a downstream phase so the runtime contract and compliance posture stay in scope.
  3. Open the matching local instruction document for the likely next stage and use it as the detailed source of truth.
  4. Confirm the required inputs for that path:
    • For description, verify the work creates a new generated skill or changes the generated skill's user-facing contract.
    • For scaffold, verify the name is present, the approved description contract exists, and the work is intended to become a Rust CLI Skill project.
    • For extend, verify both project_path and feature.
    • For validate, verify project_path.
    • For publish, determine whether the requested channel is repo_native, npm, or both. Load ../cli-forge-publish/planning-brief.md for repo-native work and ../cli-forge-publish-npm/planning-brief.md for npm work, verify whether the user wants readiness review, dry-run, or live side effects, and whether the work should start with validation or can proceed directly to the matching publish child skill.
  5. Inspect the filesystem when it helps disambiguate the stage:
    • missing target directory usually means description followed by scaffold
    • existing scaffolded project plus feature request means extend
    • existing generated skill plus description-impacting request means description followed by extend or validate
    • explicit audit or post-change verification means validate
    • release, repo-native dry-run, rehearsal, destination-config work, or final post-validation closure means publish
    • npm publication, npm package-set review, or combined repo-native plus npm distribution still begins as publish, but the downstream child skill must be selected explicitly
  6. Route to the next stage skill and carry forward the resolved inputs.

Classification Examples

  • Create request with no project directory yet -> route to description.
  • Existing generated skill plus purpose/positioning change -> route to description.
  • Existing scaffolded project plus stream or repl request -> route to extend.
  • Audit or post-change verification request -> route to validate.
  • Successful validation with no explicit release action -> route to publish in report_only mode.
  • Publish the shipped CLI to npm -> route to publish-npm after confirming validation status and target-repository context.
  • Mixed implementation and release request with unresolved project state -> return to the earliest incomplete stage instead of skipping ahead to publish.

Done Condition

This stage is complete only when one downstream phase is clearly selected and the next stage has the inputs it needs.

Next Step

Related skills
Installs
13
First Seen
Apr 2, 2026