tine replay

Replay the steps of a saved run, printing each step's label in order. This is useful for reviewing what an agent did without re-executing anything — it reads from the saved .tine file and prints the step sequence.

Usage

Terminal
tine replay <run_id> [--from-step <index>]

Arguments

  • <run_id> — The run to replay. Accepts a run ID prefix or a .tine file path.
  • --from-step <index> — Optional. Start replaying from this step index instead of the beginning.

Basic Replay

Terminal
# Replay all steps from the beginning
tine replay a3f8c1d2

# Replay starting from step 5
tine replay a3f8c1d2 --from-step 5

Full replay output:

Terminal
Replaying run_a3f8c1d2 (12 steps)...

Step  0  [think]  Planning research approach
Step  1  [tool]   search("solid-state battery 2026")
Step  2  [model]  Analyzing 8 search results
Step  3  [tool]   fetch("https://example.com/review")
Step  4  [tool]   fetch("https://example.com/update")
Step  5  [model]  Cross-referencing sources
Step  6  [think]  Evaluating conflicting claims
Step  7  [model]  Synthesizing findings
Step  8  [tool]   fetch("https://example.com/specs")
Step  9  [model]  Final synthesis
Step 10  [tool]   write("battery_report.md")
Step 11  [done]   Report written to battery_report.md

Replay complete.

Partial Replay

Use --from-step to skip the early steps and focus on a specific section of the run. This is helpful when debugging a long run where you already know the first N steps were correct.

Terminal
Replaying run_a3f8c1d2 from step 5 (7 steps)...

Step  5  [model]  Cross-referencing sources
Step  6  [think]  Evaluating conflicting claims
Step  7  [model]  Synthesizing findings
Step  8  [tool]   fetch("https://example.com/specs")
Step  9  [model]  Final synthesis
Step 10  [tool]   write("battery_report.md")
Step 11  [done]   Report written to battery_report.md

Replay complete.

When to Use Replay

  • Reviewing past runs: Quickly scan what an agent did without opening the .tine file manually.
  • Debugging: Walk through the step sequence to identify where the agent went off track, then use tine fork to branch from that point.
  • Documentation: Capture the replay output to include in reports or logs showing exactly what the agent did.