fix(analysis): make /ask reliable in real CLI Bridge sessions

[drewstone]

Summary

• Front-load the /ask output contract so real analysts return valid cited findings.
• Wait for visible terminal state in the live demo and packed-package flows.
• Replace two stale demos with one current CLI Bridge workflow.
• Remove duplicate UI captures, stale work logs, and a 7 MB generated event dump.
• Release the result as @tangle-network/braid@0.1.3.

Direct product proof

The recording uses the packed 0.1.3 package through Local CLI Bridge and Claude Code.
The coding run completed 15 of 15 workspace tests.
The /ask run returned 5 of 5 citation-supported findings in two model calls.
The captured model cost was $0.4025358, with 54.171 seconds wall time.

Verification

• pnpm test passed.
• pnpm typecheck passed.
• pnpm format:check passed across 745 files.
• pnpm check:release passed all 154 requirement entries.
• The exact packed package passed CLI, RPC, terminal, plain-text, accessibility, retry, cancellation, and shutdown flows.
• Gitleaks found no leaks in the four commits.
• The branch merges cleanly with current main.

Upstream findings

• fix(analyst): keep Python and TypeScript finding validation identical agent-eval#606 tracks the Python and TypeScript finding-schema mismatch exposed by this run.
• feat(runtime): relay live executor output through streamAgentTurn agent-runtime#762 tracks missing normalized tool results in runtime telemetry.