Proposal: use sc-compose to author beads formulas (automate complex release/research workflows)

[randlee]

Proposal: use sc-compose to author beads formulas (automate complex release/research workflows)

Context

The SC ecosystem runs recurring multi-step agent workflows through bd formulas — TOML workflow templates that pour into bead DAGs (see sc-release.formula.toml: press-release → contessa docs-review → repowise → publish-verify → announce, and the new per-repo atm-core-release-research formula: one serial research bead per audience persona).

These formulas are currently hand-written, and the bd formula engine has a hard limitation that makes that painful:

• LoopSpec iterates over integers only (count / until / range), exposing the index as a var. There is no foreach-over-a-list primitive.
• VarDef supports only string/int/bool — no list type.

So "generate N persona beads from a persona list" can't be expressed as a loop; we have to hand-expand each persona into its own [[steps]] block (5 near-identical blocks per repo, × 15 repos).

The opportunity

sc-compose is already the SC composition engine: fragments + declared variables + multi-pass rendering, resolved across runtimes. The same mechanism could compose beads formulas — emit .formula.toml from shared fragments, closing the loop primitive gap at a higher level.

Concepts to explore (this is the ask)

1. A formula-composition render target. A --mode formula (or --kind formula) that renders bd formula TOML from fragments + vars, the way --mode profile renders agent profiles. Output lands on the bd formula search path (or a declared path).

2. List-typed variables + foreach expansion. Add a list var type (e.g. personas = ["Agent Orchestrator", ...]) and a foreach/expand construct that emits one step per element — closing the LoopSpec integer-only gap. A persona-research fragment parameterized by (persona story, git-audit template, closure contract) would then expand to N serial steps.

3. Fragment library. Versioned, composable fragments for the recurring shapes: git-audit-research (dual-pass review + grounding + output template), press-release-per-persona, closure-contract. Shared across all 15 repos instead of copy-pasted per repo.

4. Colocate formula source with the persona SSOT. Formula source lives next to hendrix/alpha-prime/<repo>/user-stories.md; sc-compose renders it into the bd search path. Personas and the formula that audits them change together.

Acceptance sketch

Authoring a per-repo research formula should become:

sc-compose render --mode formula --kind release-research \
  --repo atm-core --personas "Agent Orchestrator,Agent (agent-first),..." \
  --var version --var parent-bead

…instead of hand-writing 5 serial [[steps]] blocks. The emitted formula must pour a valid bd DAG (dry-run passes) and honor the closure contract (report bead with parent, back-reference, close-on-evidence).

Not in scope (for now)

• Changing bd/beads itself — this is about sc-compose emitting what bd already consumes.
• Full runtime execution of formulas; only authoring/generation.

---

Filed by alpha-prime (harness engineering PM). Discussion welcome — this is a concept proposal, not a committed design.