Command: summary¶
agent-trace summary <ACTION> ...
Purpose: Configure pluggable session-end summarization: an arbitrary command (or built-in preset) reads the raw transcript on stdin and prints a short summary on stdout. agent-trace stores stdout keyed by conversation_url.
ACTION is required.
summary enable¶
agent-trace summary enable --command CMD [--timeout SECONDS]
| Option | Dest / meaning | Required | Default | Purpose |
|---|---|---|---|---|
--command |
CMD string |
yes | — | Shell-invoked executable: stdin = transcript bytes, stdout = summary text. |
--timeout |
seconds | no | 30 in typical defaults when omitted | Subprocess wall-clock limit; must be positive int when provided. |
Project requirement: Initialized project.
Side effects: Sets summary.enabled + summary.command in project-config.json (and timeout if supplied).
summary disable¶
agent-trace summary disable
Turns off summarization (summary.enabled false).
summary presets¶
agent-trace summary presets
Lists built-in preset aliases with descriptions (local CLI wrappers).
summary use¶
agent-trace summary use <preset_alias> [--model MODEL] [--timeout SECONDS]
| Positional | Allowed values |
|---|---|
preset_alias |
claude-summary, cursor-summary, ollama-summary |
| Option | Applies to | Purpose |
|---|---|---|
--model |
primarily ollama-summary |
Model name passed to ollama run (ignored by other presets). |
--timeout |
all | Same as enable. |
Effect: Writes a summary.command that invokes an internal summary preset-run … indirection so the stdin/stdout contract stays uniform.
Preset behaviors (high level):
| Alias | External tool | Transport notes |
|---|---|---|
claude-summary |
claude -p "<prompt>" |
Prompt may be large; OS argv limits can bite huge transcripts. |
cursor-summary |
cursor agent … |
Prompt on stdin to avoid argv limits (with --trust style flags per implementation). |
ollama-summary |
ollama run <model> |
Prompt on stdin; default model llama3.1:8b when --model omitted. |
summary generate¶
agent-trace summary generate [--conversation-url URL] [--session-id ID]
Purpose: Re-run summarization manually for:
- A single
conversation_url(usuallyfile:///…transcript), or - Every URL referenced by traces in a
session_id/conversation_id.
At least one of --conversation-url or --session-id is required; if both are omitted the command exits 1 with a usage message.
summary show¶
agent-trace summary show [COMMIT]
| Positional | Default | Purpose |
|---|---|---|
COMMIT |
HEAD |
Show {conversation_url: summary} map derived for that commit. |
Internal: summary preset-run¶
This action exists for preset wiring and is hidden from --help (argparse.SUPPRESS). Humans should use summary use. Automation authors should not depend on its stability.
agent-trace summary preset-run <preset_alias> [--model MODEL]
Storage & downstream consumption¶
Summaries live in session-summaries.jsonl. Downstream: blame / context / git notes builders may incorporate summaries when configured.
Related¶
- Session summaries concept
- summary_presets.py source for exact subprocess argv.