Runtime local: run one second-brain workflow locally

[tnez]

Parent: #103

Related: #100

Depends on: runtime package discovery/delegation (#107), local runtime package skeleton (#108)

Context

The core product gap is not just that runtime packages exist. The second-brain example needs to become runnable.

This issue is the first end-to-end acceptance target for the runtime effort.

Goal

Make one target from examples/second-brain run through the local runtime and produce a durable Agentic result.

Expected shape:

cd examples/second-brain
agentic runtime add local
agentic runtime init local
agentic runtime run <target>

The exact target can be chosen during implementation. Prefer something concrete and small, such as an inbox review, project review, or kickoff workflow.

What counts as "runnable"

A successful run should demonstrate at least:

• The local runtime can load the Agentic workspace.
• The runtime can select or receive a target.
• A persona/skill/workflow path is visible in the run.
• The run produces a durable Agentic artifact or invocation record.
• The user can inspect what happened afterward.

Non-goals

• Do not require Cloudflare.
• Do not require daemon/service mode.
• Do not require scheduling.
• Do not require external artifact storage.
• Do not require a perfect generalized workflow evaluator if the MVP can be smaller.

Acceptance

• A documented command sequence runs one second-brain target locally.
• The result is durable in Agentic terms, not just terminal output.
• The command sequence is suitable for the example README.
• Any missing polish or generalization is filed as follow-up issues.

[dottie-weaver]

Dogfood verified in session 2026-06-15T20-53-heimdall-agentic-111-dogfood.md.

Full research-loop workflow ran end-to-end:

• 6/6 workflow nodes completed (frame-question → gather-sources → write-brief → validate-taxonomy → finalize-brief → decide-next-actions)
• Finalized research-brief artifact with valid para:area/agentic-storage-research tag
• Two follow-up tasks created by the decide-next-actions node
• All artifacts durable on filesystem (not just terminal output)
• Run record persisted

Acceptance criteria met:
✅ Documented command sequence runs one second-brain target locally
✅ Result is durable in Agentic terms, not just terminal output
✅ Command sequence suitable for example README
✅ Missing polish filed as follow-ups (#112)

Note: seed tasks on main don't exist yet (they're on the PR #102 branch), so a manual task creation step was needed before task next. This should be noted in #112 docs or resolved when #102 merges.