Skip to content

evidence: claude_code 2.1.270 — re-witness wanted #148

Description

@github-actions

claude_code has moved since its evidence was taken:

  • pre_tool: evidence taken against 2.1.263 (7 days ago); 2.1.270 is now published.
  • post_tool: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • tool_failure: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • prompt_submit: evidence taken against 2.1.263 (7 days ago); 2.1.270 is now published.
  • session_start: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • session_end: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • stop: evidence taken against 2.1.263 (7 days ago); 2.1.270 is now published.
  • pre_compact: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • subagent_start: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • subagent_stop: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • instructions_loaded: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.
  • file_changed: evidence taken against 2.1.247 (17 days ago); 2.1.270 is now published.

Re-witness it

You need claude_code installed and runnable headlessly from a terminal.

pip install agentseam
python3 tools/experiment.py run --agent claude_code --event pre_tool \
    --driver "<your headless CLI invocation, containing {prompt}>" \
    --agent-version 2.1.270 --record --reporter @yourhandle
python3 tools/experiment.py run --agent claude_code --event prompt_submit \
    --driver "<your headless CLI invocation, containing {prompt}>" \
    --agent-version 2.1.270 --record --reporter @yourhandle
python3 tools/experiment.py run --agent claude_code --event stop \
    --driver "<your headless CLI invocation, containing {prompt}>" \
    --agent-version 2.1.270 --record --reporter @yourhandle

This writes data/recordings/claude_code@2.1.270.json and touches nothing else on your machine.

A run against the reference or recorded driver never counts as a re-witness --
only a real, running claude_code does.

Submit it

  • Preferred: open a PR adding that recording file plus the updated
    data/matrix.json per-claim test pointers (see any recorded row for the shape).
  • Or: open an "Evidence report" issue and paste the recording JSON for a
    maintainer to land.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions