CLI Reference¶
agentdelta diff¶
Compare two agent trace files and report any behavioral regression.
| Option | Default | Description |
|---|---|---|
--format |
rich |
Output format: rich | json | markdown |
--fork-threshold |
0.70 |
Similarity below this marks a fork point |
--match-threshold |
0.85 |
Similarity above this is a match (no change) |
--show-matches |
false |
Include unchanged steps in the output |
--exit-code |
false |
Exit 1 if a regression is detected (for CI pipelines) |
Examples¶
# Default rich terminal output
agentdelta diff baseline.jsonl candidate.jsonl
# JSON output for programmatic use
agentdelta diff baseline.jsonl candidate.jsonl --format json
# Markdown for a GitHub PR comment
agentdelta diff baseline.jsonl candidate.jsonl --format markdown > diff.md
# CI gate — fails the pipeline if traces diverged
agentdelta diff baseline.jsonl candidate.jsonl --exit-code
# Show all steps including unchanged ones
agentdelta diff baseline.jsonl candidate.jsonl --show-matches
# Tighter thresholds (stricter regression detection)
agentdelta diff baseline.jsonl candidate.jsonl \
--fork-threshold 0.80 \
--match-threshold 0.90
agentdelta inspect¶
Print a step-by-step summary of a single trace file.
No options. Outputs a table of steps with node type, content preview, and node ID.
Example¶
run_id: v1.0 · 6 nodes · 5 edges
Step Type Content
1 start What is the weather in Tokyo?
2 llm I should look up the current weather...
3 tool_call get_weather(location='Tokyo')
4 tool_return {"temp": 22, "condition": "sunny"}
5 llm The weather in Tokyo is 22°C and sunny.
6 end Tokyo: 22°C, sunny.
Exit codes¶
| Code | Meaning |
|---|---|
0 |
Success — no regression detected (or --exit-code not set) |
1 |
Regression detected (only when --exit-code is set) |
2 |
Invalid arguments or file not found |