Agent Skills: Debundle Lane Worker

Debundle Lane Worker

Use this role for one scoped implementation assignment: a seed cluster from intake, a binding-patch/residual cohort, or a firm reorganization task from the architect.

Shared CLI workflows land here for binding moves, renames, module merges, and atom-split recovery:

@references/docs/cli.md @references/docs/selectors.md @references/docs/spec_editing.md

Read other bundled references as needed:

• references/workflow.md for role boundaries and failure routing
• references/README.md for the crate pitch + Comments
• references/module_shape.md for seam and layer-ownership heuristics

Inputs

The orchestrator or project adapter provides:

• worktree root and expected base SHA
• assignment with owner IDs, binding IDs, proposed destination, and notes
• <graph>, <modules-dir>, <emitted-js-root>, and optional source root
• project conventions/taxonomy docs
• exact gate, regen, uniqueness-check, and commit expectations

Procedure

1. Confirm the worktree is at the expected base before editing.

2. Check the assignment against the current graph with debundle describe and debundle show-source; if needed, scan with debundle atoms, coverage, or cluster <sym> (see the shared CLI guide above).

3. Read each binding's surrounding code: consumers, dependencies, and nearby implementation details.

4. Choose a module boundary that looks like a real JavaScript seam under the project conventions.

5. Apply the assignment using the shared guide's CLI workflows. Prefer bindings assign, bindings rename, and modules merge over hand-editing module YAML. Which proposals bindings assign --batch takes directly, and what the others need: references/docs/cli.md § --batch JSON format.

   For selector-stabilization assignments, follow references/docs/selectors.md (the ladder and § "Bulk conversion loop"): confirm the bucket with selector-debt, draft with synthesize-selectors dry-run scoped to the assignment, minimize before --apply, then run git diff --check, the adapter's gate/regen command and selector-debt again to report the debt delta. Never modify the upstream/source bundle. A binding Ducktape cannot yet stabilize becomes selector debt (references/docs/selectors.md § Selector debt), routed back to Ducktape tooling.

6. Remove now-owned entries from the non-emitting rename/annotation patch stream when the project uses one.

7. Run the adapter-provided uniqueness check, gate, and regen commands.

8. Commit one reviewable branch and report the result.

Boundary Heuristics

A good module has a coherent reason to exist: stable public surface, internal references dominating external references, clear layer ownership, or multiple meaningful consumers. Member count alone is not the rule.

Tiny modules are a smell — try to fold them. The chunker over-splits when it emits a separate module for what a developer would have written inline in a larger file. Judge by LINES OF CODE, not member count: a one-binding module that is a 500-line React component is idiomatic and must be left alone; the smell is small-LOC standalone files (a 1-3 line accessor, predicate, constant, or wrapper). Fold a small-LOC module into its single real consumer (excluding non-semantic re-export catalogs / bundle barrels), or into a sibling that was clearly the same original source file (use source_location adjacency / shared CSS-module class prefixes as evidence). Do NOT fold widely-consumed shared primitives (a shared constant, a React context, a public predicate), real public-API/service/class boundaries, or anything whose fold would cross a layer boundary or break the gate. See references/module_shape.md.

Usually avoid standalone modules for:

• primitive constants with one consumer
• one-line helpers with one consumer
• enum-like values that only parametrize a larger owner
• local style/config/data artifacts with no public contract

Do not co-locate solely by consumer count when that would violate layer ownership. Policy, domain, persistence, infra, and integration logic keep their own homes even when a presenter is currently the only caller.

You may expand the batch when the assigned peel would split an atomic unit or create a worse module shape. You may also skip assigned items when their natural owner still belongs to a larger atomic unit that should move together.

Failure Handling

• If the graph is stale, rerun the adapter-provided graph refresh or report the stale evidence.
• If bindings assign refuses with an atom-split diagnostic, follow the shared guide's atom-split workflow before changing the assignment shape.
• If the gate rejects via debundle run, read the structured cycle/report output (cycles.json, atomic_unit_conflicts.json) first. Use the cut/evidence if present.
• If broad minified-source analysis is needed, stop and ask for intake grounding.
• If the destination is architecturally unclear, stop and route to the architect instead of inventing a dump bucket.

Report

Keep the report short:

• branch and commit hash
• gate and regen result
• destinations changed and bindings moved
• skipped proposals or units and why
• architecture or intake follow-ups discovered