[chaliy]

Requested by Михайло · project thread

What changed

examples/pydantic_ai_bash_agent.py prints result.output instead of result.data.

Why

pydantic-ai 1.0 removed AgentRunResult.data. bashkit[pydantic-ai] requires pydantic-ai>=1.0, so the example crashed right after a successful agent run. Found during the pre-release integration pass (no CI job runs this LLM-backed example).

Before / After

Before (pydantic-ai 2.66.0, local 0.20.3+main wheel, real Anthropic key):

  File "examples/pydantic_ai_bash_agent.py", line 71, in main
    print(result.data)
AttributeError: 'AgentRunResult' object has no attribute 'data'

After: exit 0, agent summary printed (FizzBuzz script created, shown, sourced, correct 1..20 output).

Risk

• Low
• Example only, no library code.

Checklist

• Tests added or updated (example run manually against live model)
• Backward compatibility considered (extra already requires pydantic-ai>=1.0)

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	bashkit	0ed3485	Commit Preview URL Branch Preview URL	Nov 10 2026, 08:13 PM

[bashkit-pr-reviewer]

One finding only; the example fix itself looks correct.

[bashkit-pr-reviewer]

Review summary (head 0ed3485)

Verdict: Clean one-line example fix; no findings. Safe to merge.

Findings: None on this push. The earlier P2 (unrelated-cargo-lock-churn, 52 lines of incidental stacker/psm/object additions to crates/bashkit-python/test-fixtures/random-fs/Cargo.lock) is resolved — the lockfile is no longer in the diff.

Context checked beyond the diff: full diff visible (truncated: false, now only examples/pydantic_ai_bash_agent.py line 71: result.data → result.output); confirmed against the checked-out head plus the previously read crates/bashkit-python/pyproject.toml extra (pydantic-ai>=1.0, consistent with .data being gone) and CI (examples job covers Rust examples only, so the PR's manual live-model run is the right verification for this LLM-keyed example).

Not reviewed: nothing outstanding; the single changed file was fully visible.