Feature request: debug/inspect mode for flow execution

[ameet]

Problem

When a step fails in a long flow (e.g., step 27 of 40), debugging requires:

1. Re-running the entire flow from step 1
2. Reading through all step logs to reconstruct intermediate state
3. Guessing at what the resolved interpolations looked like at runtime

There's no way to:

• Stop execution at a specific step
• Inspect the resolved value of $.steps.X.output at any point
• Dry-run a flow to see what interpolations resolve to without executing

Proposal

Three complementary capabilities:

1. Stop-at flag:

one flow execute my-flow --stop-after=step15

Executes steps 1-15, prints step 15's output, and exits. Useful for isolating where things break.

2. Inspect resolved interpolations:

one flow execute my-flow --dry-run

Resolves all {{$.steps.X.output.Y}} and $.input.Z references, prints the resolved values per step, but doesn't execute any steps. Catches wiring bugs (wrong field names, missing null guards) without burning API credits.

3. Step output dump:

one flow execute my-flow --dump-steps=/tmp/flow-debug.json

Writes all step outputs to a file during normal execution. Enables post-mortem debugging without re-running.

Why this matters

Flow debugging time scales linearly with flow length. A 40-step flow with a bug at step 30 currently costs the full execution time of steps 1-29 on every debug attempt. These features would reduce that to near-zero for wiring bugs (dry-run) and one-shot for runtime bugs (stop-at + dump).