add IBM Bob agent adapter

[fredkwok-it]

Summary

Add support for IBM Bob Shell CLI as a new coding agent adapter.

Changes

• next/src/lib/agents/detect.ts: Add bob agent definition with IBM vendor info and fallback models
• next/src/lib/agents/argv.ts:
  • Add bob case in buildArgv() to use --output-format stream-json
  • Add bob parsing logic in parseLineWithState() to handle text/content/message fields
• next/src/lib/agents/invoke.ts: Add prompt echo suppression to avoid duplicating system prompts in output
• README.md: Update agent count badge (8 → 9 CLIs) and add IBM Bob to supported agents table
• README.zh-CN.md: Update Chinese README with same changes

Testing

• ✅ Bob CLI is installed and available at /opt/homebrew/bin/bob
• ✅ Verified bob supports --output-format stream-json flag
• ✅ Integration follows the same pattern as existing agents (qwen, opencode)

Implementation Details

Bob is invoked via stdin with stream-json output format, similar to other agents. The adapter reuses the existing JSON-line parser infrastructure and handles three common output fields: text, content, and message.

The implementation includes a workaround in invoke.ts to suppress prompt echo that bob may send back as the first streamed delta, preventing system prompt pollution in user-facing output.

Checklist

• Agent adapter added to detect.ts
• Command-line arguments configured in buildArgv()
• Stream parser added in parseLineWithState()
• README.md updated with agent entry
• README.zh-CN.md updated with agent entry
• Changes follow existing agent adapter patterns
• No new dependencies added

References

• IBM Bob Shell: https://www.ibm.com/products/watsonx-code-assistant/bob
• Follows contribution guidelines from CONTRIBUTING.md

[nettee]

I found one blocking issue in the new Bob adapter. The inline comment has the concrete fix; the rest of the changed ranges looked consistent with the existing agent wiring.

_{🔁 Powered by Looper · runner=reviewer · agent=codex · An autonomous AI dev team for your GitHub repos.}

[nettee]

IBM's Bob Shell docs say non-interactive output includes both the final answer and intermediary thinking by default, and they expose --hide-intermediary-output specifically to limit stdout to the final completion output. This adapter only adds --output-format stream-json, while the new Bob parser below forwards every top-level text / content / message string into the preview. In practice that means Bob's reasoning and tool chatter can be appended straight into the generated HTML or markdown, which breaks the user-visible document instead of just adding metadata. Please either add --hide-intermediary-output here or tighten the Bob parser to emit only final assistant content, and add a regression test in next/src/lib/agents/__tests__/argv.test.ts that covers a "thinking + final answer" Bob stream.

_{🔁 Powered by Looper · runner=reviewer · agent=codex · An autonomous AI dev team for your GitHub repos.}

[fredkwok-it]

Thanks for the review! I've addressed the feedback:

✅ Added --hide-intermediary-output flag to the Bob adapter in buildArgv() to suppress thinking/reasoning output
✅ Added regression tests in argv.test.ts covering Bob's stream-json parsing, including a test case for the final answer scenario
✅ Updated both READMEs with the correct invocation command

The flag ensures that only the final completion output is sent to stdout, preventing Bob's intermediary thinking from polluting the generated HTML/markdown artifacts.

All tests pass locally (pnpm -F @html-anything/next test).

Ready for re-review!

[nettee]

@fredkwok-it I rechecked the Bob follow-up commit against the changed ranges. The adapter now hides intermediary output, the supported-agent/docs updates are consistent, and the Bob parser coverage added here matches the current adapter shape. Nice follow-through on the review feedback.

_{🔁 Powered by Looper · runner=reviewer · agent=codex · An autonomous AI dev team for your GitHub repos.}