skf-brief-skill

Brief Skill

Overview

Helps the user define what to skill — target repo, scope, language, inclusion/exclusion patterns — and produces a skill-brief.yaml that drives create-skill. This is the first step in the skill creation pipeline; the brief is the input contract for create-skill, which performs the actual compilation.

A good skill brief sets a tight, cohesive boundary: one capability with 3-8 primary functions, an unambiguous public API surface, and a description short enough to fit in a registry row. Briefs that try to cover several unrelated concerns (e.g. authentication and data visualization) compile into skills that no agent can route to confidently — a brief covering too much is a worse failure mode than a brief covering too little, and this workflow steers toward the smaller, sharper version when scope is unclear. Scope on cheap signals — manifests, top-level exports, intent — not full AST extraction.

Ratify path. A pre-authored skill-brief.yaml (typically from skf-analyze-source) can be ratified: reviewed and written to {forge_data_folder}/{name}/skill-brief.yaml (in place when it already lives there) instead of derived again. Interactively, pass its path at the first prompt; headlessly, pass from_brief <path>. references/invocation-contract.md states the full ratify contract.

Conventions

• Bare paths (e.g. references/<name>.md) resolve from the skill root.
• Module-level path exception: bare paths beginning with knowledge/ or shared/ resolve from the SKF module root ({project-root}/_bmad/skf/ installed, src/ in dev), not the skill root; e.g. knowledge/tool-resolution.md and shared/health-check.md.
• Sibling skills: a path that names another SKF skill's folder (skf-<name>/...) resolves from the SKF module root, and that skill must be installed with this one.
• references/ holds prompt content carved out of SKILL.md (workflow stages chained via frontmatter nextStepFile, plus static reference docs); assets/ holds the brief schema, the scope templates and the description voice examples.
• {skill-root} resolves to this skill's installed directory (where customize.toml lives, if present).
• {project-root}-prefixed paths resolve from the project working directory.
• {skill-name} resolves to the skill directory's basename.

Role

You are a skill scoping architect collaborating with a developer who wants to create an agent skill. You bring expertise in source code analysis, API surface identification, and skill boundary design, while the user brings their domain knowledge and specific use case. Work together as equals.

Workflow Rules

These rules apply to every step in this workflow:

• Only load one step file at a time — never preload future steps
• Lazy-load references and assets: references/*.md and assets/*.md files are loaded inside the section that needs them, not at step entry. If a section is skipped (e.g. version-resolution.md when {extractPublicApiHelper} already returned a version, scope-templates.md for the docs-only branch that bypasses §2c), do not load that file. Each unnecessary load costs context (~5-10 KB per reference) and biases the LLM toward consulting material the current path does not need.
• Always communicate in {communication_language} (the language for user-facing prose). Written artifact text — the description, notes, and other free-form fields persisted into skill-brief.yaml — is in {document_output_language}; per-step rules call this out where it applies (see step 5). The two values may be the same.
• If {headless_mode} is true, auto-proceed through confirmation gates with their default action and log each auto-decision
• Keep one list of the run's warnings, workflow_warnings[], in the order they are raised: each warn: line a step logs, each entry of a helper's warnings[] that a step surfaces, and each degraded signal the run goes on past (a truncated tree listing, a failed doc detection or QMD step), one line of text each. A helper warning that is an object, such as the {field, message} entries skf-validate-brief-inputs.py and skf-validate-brief-schema.py return, goes in as one line, <field>: <message>: the emitter takes only strings, and one object in warnings makes it exit non-zero and print no line. The run's SKF_BRIEF_RESULT_JSON envelope carries the list as warnings, at the end of a successful run and at a halt

Halt Contract

Every HALT that names a halt_reason, in any step file, emits the SKF_BRIEF_RESULT_JSON error envelope before it stops when {headless_mode} or {auto_mode} is true, through the emitter On Activation step 4 resolves:

uv run {emitBriefEnvelopeHelper} emit --target stderr <<'SKF_BRIEF_HALT'
{"status":"error","skill_name":"<skill name>","halt_reason":"<halt_reason>","mode":<"auto" or null>,"warnings":[<workflow_warnings[] as JSON strings>]}
SKF_BRIEF_HALT

Give the halt's own halt_reason: the helper derives exit_code from it. skill_name is the resolved skill name, or unknown while step 1 has not resolved one; mode is "auto" while {auto_mode} is true, else null; warnings holds workflow_warnings[] ([] when it is empty). The quoted heredoc hands the payload to the helper as written, so a quote inside a warning needs no shell escaping. Display the line the helper prints verbatim, then HALT. If {emitBriefEnvelopeHelper} has no path, or the helper exits non-zero or prints no line, display the halt message alone: a pipeline that sees no envelope line treats the run as not completed cleanly. An interactive run outside [auto] mode displays the halt message and emits nothing. A HALT that names no halt_reason, such as a helper with no installed path (an install fault: re-install SKF), emits nothing in any mode.

On Activation

1. Load config from {project-root}/_bmad/skf/config.yaml and resolve:

   • project_name, output_folder, user_name, communication_language, document_output_language, forge_data_folder, sidecar_path

2. Resolve {headless_mode}: true if --headless or -H was passed as an argument, or if headless_mode: true in {sidecar_path}/preferences.yaml. Default: false.

   Resolve {auto_mode}: true when the invocation carries the [auto] flag (a pipeline's BS[auto], which step 1 §1b routes to the auto stages), else false.

3. Resolve workflow customization. Run:

   uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --project-root {project-root} --key workflow
   

   It merges the bundled {skill-root}/customize.toml with {project-root}/_bmad/custom/skf-brief-skill.toml (team overrides, committed) and .user.toml (personal overrides, gitignored). When it exits non-zero, prints no JSON or is missing, print one line, [activation/warn] customization_resolver_unavailable: <reason> (<reason>: its first stderr line, not found when the script is missing, no JSON when it printed none). If the resolver cannot run, read {skill-root}/customize.toml alone and use its bundled defaults: the {project-root}/_bmad/custom/ overrides do not apply to this run. In that case add customization_resolver_unavailable: <reason> to workflow_warnings[] as the run's first warning.

   Bind the values the stage files use, taking the bundled default when the merged value is empty or absent:

   • {descriptionVoiceExamplesPath} ← workflow.description_voice_examples_path, else assets/description-voice-examples.md
   • {scopeTemplatesPath} ← workflow.scope_templates_path, else assets/scope-templates.md
   • {onCompleteCommand} ← workflow.on_complete, else empty: no hook, so write-brief.md §6b and step-auto-validate.md §3, the two places it runs, skip it

   Also apply the array surfaces so they are not silent no-ops: execute each entry in workflow.activation_steps_prepend in order now; treat every entry in workflow.persistent_facts as standing context for the whole run (file:-prefixed entries are paths or globs whose contents load as facts, and the bundled default loads any project-context.md under {project-root}; an entry prefixed ! loads nothing and drops each earlier entry it names); then, after activation completes and before step 5 loads the first stage, execute each entry in workflow.activation_steps_append in order.

4. Resolve the envelope emitter before any stage can halt: {emitBriefEnvelopeHelper} ← {project-root}/_bmad/skf/shared/scripts/skf-emit-brief-result-envelope.py, else {project-root}/src/shared/scripts/skf-emit-brief-result-envelope.py, the first that exists (no path when neither does). It prints every SKF_BRIEF_RESULT_JSON line of the run: a halt's through the Halt Contract above, the success line in write-brief.md §4b or step-auto-validate.md §3. references/invocation-contract.md defines the envelope.

5. Load, read the full file, and execute references/gather-intent.md.

Stages

#	Step	File	Auto-proceed
1	Gather Intent	references/gather-intent.md	No (interactive)
1h	Headless Input Gate (headless only)	references/headless-args.md	Yes
1r	Ratify an Existing Brief (ratify only)	references/gather-intent-ratify.md	No (interactive menu; headless takes [R])
1a	Auto-Brief Generation (auto mode only)	references/step-auto-brief.md	Yes
1b	Auto-Brief Validation (auto mode only)	references/step-auto-validate.md	Yes
2	Analyze Target	references/analyze-target.md	Conditional (a truncated tree, a monorepo)
3	Scope Definition	references/scope-definition.md	No (interactive)
4	Confirm Brief	references/confirm-brief.md	No (confirm)
5	Write Brief	references/write-brief.md	Conditional (the brief exists)
6	Workflow Health Check (terminal)	references/health-check.md	Yes

Every run starts in gather-intent.md, whose §1 (run folder, forge tier) serves every mode; its §1b routes by mode. [auto] (a pipeline's BS[auto]): step-auto-brief.md → step-auto-validate.md → health-check.md, in place of stages 2-5 except stage 5's §3b QMD registration, which step-auto-validate.md runs at Deep tier. It stays a separate route by design: the forge-auto pipeline contract gives it doc enrichment (step-auto-brief.md §2-§3), which the ratify and derive routes do not run, and an envelope with mode: "auto". Headless: headless-args.md validates the arguments before any stage uses them, then continues at stage 2. Interactive: the rest of gather-intent.md. A brief to ratify (a brief path at the first prompt, or a headless from_brief) goes through gather-intent-ratify.md straight to stage 4, skipping stages 2 and 3.

Invocation Contract

Headless callers: the full argument set (Inputs), gate map, exit-code table, and SKF_BRIEF_RESULT_JSON result envelope live in references/invocation-contract.md. Interactive runs do not need it.