Skip to content

Conversation

@christso
Copy link
Collaborator

Summary

  • Add dedicated pi-coding-agent provider for the Pi Coding Agent CLI from pi-mono
  • Support file attachments using Pi's native @path syntax
  • Extract tool trajectory/traces from Pi's JSONL output for evaluation
  • Display log file paths in console output during eval runs (similar to Codex)

Changes

  • New PiCodingAgentProvider class with JSONL stream parsing
  • New pi-log-tracker.ts for real-time log path notifications
  • Escape @[Role]: markers in prompts to avoid Pi interpreting them as file paths
  • Add example target configuration in examples/features/.agentv/targets.yaml

Test plan

  • Build passes
  • Functional test with bun agentv eval examples/features/evals/basic/example-eval.yaml --target pi_coding_agent
  • Log paths displayed in console during eval run
  • Tool calls extracted into outputMessages for trace evaluation

🤖 Generated with Claude Code

Add dedicated provider for pi-coding-agent CLI from pi-mono repository.
This enables AgentV to evaluate the Pi autonomous coding agent with
full JSONL output parsing, tool call extraction, and stream logging.

- Add PiCodingAgentProvider class with JSONL event parsing
- Add 'pi-coding-agent' to ProviderKind with 'pi' alias
- Add PiCodingAgentResolvedConfig type and resolver
- Support provider, model, tools, thinking level configuration
- Add example target configuration and test eval
Use pi CLI's native @path syntax for file attachments instead of
embedding file:// URIs in the prompt text.
…icts

Pi CLI interprets @ as file prefix, but AgentV prompts use @[Role]: markers
for multi-turn conversations. Escape these patterns to [[Role]]: format.

Also removes pi-agent specific evals - use basic evals which are agent-agnostic.
Add log path tracking for Pi Coding Agent similar to Codex CLI.
Log file paths are now displayed in the console during eval runs
with a "Pi Coding Agent logs:" header.
Add log_format config option (default: 'summary', or 'json' for raw events).
This matches Codex CLI behavior and allows full output preservation.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants