Function buildPlaybookRunBootstrapPrompt

internal/app/do.go:777–814

buildPlaybookRunBootstrapPrompt is the prompt for playbook-run tasks. Adds an explicit `flow show playbook ` context-load step and frames the run's brief as an authoritative snapshot — the session must execute against that snapshot, not re-read the live playbook brief (which may drift between

(runSlug, playbookSlug string, isFirstRun bool)

// quotes, backticks, or dollar signs — because it gets shell-quoted
// as a single positional argument to `claude`.
//
// The session's UUID is pre-allocated by `flow do` and passed via
// `claude --session-id <uuid>`, so there is no self-registration step
// here. The session loads context in order: task brief + task updates,
// then (if any) project brief + project updates, then CLAUDE.md files
// in the work_dir. The flow skill enforces this sequence too; the
// bootstrap prompt is a backup in case the skill isn't auto-activated.
// Kept for callers (and tests) that don't track first-run state. New
// callers should use buildBootstrapPromptForKindV2 to opt into the
// first-run variant when relevant.
func buildBootstrapPromptForKind(slug, kind, playbookSlug string) string {
	return buildBootstrapPromptForKindV2(slug, kind, playbookSlug, false)
}

// buildBootstrapPromptForKindV2 is the kind-aware dispatcher with first-
// run awareness for playbook runs. When isFirstRun=true on a playbook
// run, a richer "capture-aggressive" prompt is emitted that nudges the
// session to harvest scripts, edge cases, and decision rules back into
// the live playbook brief / sidecar files.
func buildBootstrapPromptForKindV2(slug, kind, playbookSlug string, isFirstRun bool) string {
	if kind == "playbook_run" {
		return buildPlaybookRunBootstrapPrompt(slug, playbookSlug, isFirstRun)
	}
	return buildTaskBootstrapPrompt(slug)
}

// buildTaskBootstrapPrompt is the prompt for regular tasks.
func buildTaskBootstrapPrompt(slug string) string {
	return fmt.Sprintf(
		"You are the execution session for flow task %s. Do ALL of the following in order before touching code:\n"+
			"1. Invoke the flow skill via the Skill tool. This loads the operating manual that governs how this session works: workflows, bootstrap contract, KB discipline, and scope-creep detection.\n"+
			"2. Run: flow show task. Read the file at the brief: path AND every file listed under updates:. Files listed under other: are sidecar references — load on demand when relevant, not eagerly.\n"+
			"3. If a project is listed on the task, run: flow show project <that-project-slug>. Read its brief AND every file under updates:. Files under other: are on-demand references.\n"+
			"4. Read CLAUDE.md in your work_dir and any nested CLAUDE.md files under subdirectories you will modify. These override any assumption from the brief.\n"+
			"5. Only then begin work. If any brief section is blank or unclear, ASK — do not infer.",
		slug,
	)
}

// buildPlaybookRunBootstrapPrompt is the prompt for playbook-run tasks.
// Adds an explicit `flow show playbook <slug>` context-load step and