fix(runner): cascade-skip downstream stateful steps when phase capability gate unmet

[garvitkaushik-123]

Summary

• Track context outputs from phases skipped by an unmet requires_capability gate and stop dependent steps before dispatch, including non-stateful expect_error steps and explicit context_inputs.
• Preserve declared topology: depends_on: [], targeted unrelated dependencies, and sibling any_of branches remain independent; ordinary malformed expect_error vectors keep their existing wire behavior.
• Report capability-caused prerequisite skips as neutral not_applicable outcomes while leaving generic not_applicable, missing_tool, and prerequisite_failed coverage gaps partial.
• Add a patch changeset and regression coverage for gated/satisfied paths, implicit stateful cascades, dependency scoping, branch peers, no-dispatch behavior, and compliance bundle scoring.

Closes adcontextprotocol/adcp#7115

Test plan

• npm run build
• npm run typecheck
• npm run lint (0 errors)
• node --test --test-reporter spec test/lib/storyboard-phase-capability-cascade.test.js test/lib/compliance-bundle-results.test.js test/lib/storyboard-cascade-skip-on-skip.test.js test/lib/storyboard-runner-output-contract-v2-runner.test.js (82/82 pass)
• git diff --check
• Full library run reached 14,117 passing tests before the sandbox's ephemeral PostgreSQL endpoint disappeared; remaining live-backend tests failed/cancelled with ECONNREFUSED and are left to CI's managed services.

[aao-secretariat]

Ladon verdict: Approve

Approve — well-scoped runner bugfix (adcp#7115).

Capability-gated phases now cascade-skip dependent steps before dispatch instead of sending unresolved $context placeholders on the wire, and report as neutral not_applicable in bundle scoring. Reviewer verified no pass-count inflation, correct scoping to capability gates only, preserved any_of/depends_on topology, and no regression to intentional expect_error malformed-vector behavior. Changeset (.changeset/calm-runners-skip-capabilities.md) is present alongside the src/lib/** changes, and new behavior is covered by test/lib/storyboard-phase-capability-cascade.test.js.

No Critical/High/Medium findings. Not gated (gated_paths: false), not high-risk, no author-team gate. Falls through to row 9.

[aao-secretariat]

Ladon verdict: Approve

Approve.

Clean pass with no findings. The delta since the prior approve is a single logical change: an added description on the collection field in the synced upstream schemas/registry/registry.yaml, with src/lib/registry/types.generated.ts regenerated to mirror it verbatim. Documentation/metadata only — no code behavior, wire-shape, or witness-fidelity change. Generated file matches source; no changeset needed since the change is generated/metadata-only.

Decision-table walkthrough:

• Rows 1: no critical/high findings.
• Row 2: gated_paths is false — not triggered (review_decision is REVIEW_REQUIRED, but the deterministic gate does not apply).
• Row 3: high_risk is true but the reason is (modified), not (deleted).
• Row 4: no medium findings.
• Row 5: high_risk (modified) present, but no medium finding — not triggered.
• Row 6: prior decision was approve, not escalate.
• Row 7: no no-auto-approve team match.
• Row 8: fewer than three medium findings (zero).
• Row 9: approve.

The high-risk flag fires only because schemas/registry/registry.yaml matches schemas/registry/** as a modification; with no medium-or-higher finding on that file, the modification is presumed safe per the flag-reading guidance.