Read and follow framework/work-principle.md before taking up work. Resolve this reference against the framework root; retain the consumer CWD.

Skill: risk-followup-routing

Purpose

Dispatch each remaining_findings: entry from a chief verdict to the correct downstream destination based on its target: field. Replaces the legacy risk-followup-task step which created one task per finding (wrong granularity for the actual mix of finding types).

When to call

• Build workflow close-bookkeeping gate, after chief verdict.
• Fix workflow analogous gate, after chief verdict.
• Review workflow analogous gate, after chief verdict.

The chief is responsible for annotating every entry with target: per the 7-value enum (spec 306 §4.7 + code_review_board/verdict.md §5). This skill assumes that annotation exists; if any entry has empty target:, the skill fails with a clear error and surfaces back to the chief for re-emit.

Input

• Path to chief verdict file (e.g. docs/reviews/board/<id>-consolidated.md).
• Path to consumer's context/risk-watch.md (created if absent per skills/_protocols/risk-watch-template.md).

Process

1. Parse the chief verdict's remaining_findings: block. Validate that every entry has a target: field. On missing field: FAIL with target_missing_on_<id>.
2. Group entries by target: value:
   • spec_text: collect all into a single batch.
   • new_task: each entry filed independently.
   • watch_item: collect for risk-watch append.
   • accept: the verdict entry is the record (accepted limitation — distinct from absorb_next: no future fix implied).
   • absorb_next: the verdict entry is the record; the next review brief for a touched file names it (code_review_board/brief-and-team.md §2).
   • closes_with: <id>: validate referenced finding exists in same verdict.
   • re_review: <reviewer>: collect for re-dispatch.
3. Dispatch per group:
   • spec_text batch: if non-empty, dispatch agents/spec-text-drift-batch.md once with the full list (per spec 306 §4.8). If chief verdict is FAIL and MCA fix-pass dispatches in parallel, fire both in one tool message (the parallel-dispatch invariant is in §4.8).
   • new_task per entry: dispatch task_creation skill per entry. When called from this routing, skip the duplicate-check step of task_creation (the chief verdict already deduped at finding-level).
   • watch_item: if context/risk-watch.md doesn't exist, create from skills/_protocols/risk-watch-template.md. Append each entry verbatim under ## Entries.
   • accept: no file, no fix, no task, no trigger; the verdict entry records the known accepted limitation.
   • absorb_next: no immediate action; the verdict entry stays the record until a later change to that file closes it.
   • closes_with: assert the referenced finding exists in the same verdict. On miss: FAIL with closes_with_target_missing. On hit: no further action.
   • re_review: dispatch the named reviewer with the finding cluster as scoped focus. The re-review's verdict appends to the existing chief verdict (not a new pass). One-shot — if uncertainty persists, escalate to council per existing cross-layer-decision pattern.
4. Aggregate routing report: <N> entries routed (<count> spec_text, <count> new_task, <count> watch_item, <count> accept, <count> absorb_next, <count> closes_with, <count> re_review).

Output

DELIVERS:

• Spec-text patches applied (via spec-text-drift-batch sub-agent return, when spec_text entries exist).
• Tasks filed (via task_creation per entry).
• Risk-watch append (one or more entries).
• accept / absorb_next / closes_with: the verdict entry is the record (no file).
• Re-review dispatched for re_review entries.

DOES NOT DELIVER:

• No semantic edit of finding content. The routing applies what the chief decided; it does not re-decide.
• No task duplicate-check (delegated to chief verdict's finding-level dedup; explicit skip-flag passed to task_creation).

ENABLES:

• Bulk-task-bloat avoidance (vs the legacy risk-followup-task which filed one task per finding regardless of type).
• Mechanical traceability (every finding ends up in a known location: applied diff / task / risk-watch / verdict).

Failure modes

• Missing target: field: chief verdict didn't annotate all entries. FAIL — return to chief for re-emit.
• closes_with: <id> references non-existent finding: chief emit error. FAIL with locator.
• re_review: <reviewer> references unknown reviewer: the reviewer name is not in the framework's persona registry. FAIL with reviewer name + suggestion to use a known persona.
• spec-text-drift-batch returns PARTIAL: some entries skipped. Surface the skip reasons to user; the routing succeeds for the applied entries but flags partial-completion.

Replaces

• Legacy risk-followup-task step in build / fix / review WORKFLOW.md. The legacy step's name is preserved as a deprecated alias for one release cycle to avoid breaking in-flight runs; new dispatches use risk-followup-routing.