Catalog PyPI
PACKAGE EXPLORERView in registry

agent-trace-card

PyPI

Generate local Markdown, HTML, and JSON review cards from agent trace JSON.

agent-evaluationai-agentsincident-reviewllmopsregression-testingtraces
Versions collected1
Direct dependencies0From release 0.1.2
Last observed20 Oct, 23:34 UTC

README

Version 0.1.2 documentation

agent-trace-card

agent-trace-card turns one local agent trace JSON file into a portable review card for incident discussion, handoff, and regression planning.

At a Glance

JobSummarize an agent run as Markdown, HTML, or schema-versioned JSON.
Built forAgent developers, eval engineers, incident reviewers, and governance teams.
DifferentiatorSmall, dependency-free CLI with deterministic heuristics and a checked-in v1 schema.
ProducesGoal, outcome, tools, retries, selected identifiers, failure modes, human intervention, and regression status.

Install

The current registry release is:

python -m pip install "agent-trace-card==0.1.2"

Verified Quickstart

Run from a source checkout so the synthetic fixture is available:

agent-trace-card generate \
  --from examples/refund_trace.json \
  --format json \
  --out /tmp/refund-agent-card.json

agent-trace-card validate /tmp/refund-agent-card.json

The bundled refund fixture produces a valid agent-trace-card/v1 document with outcome failed and retry count 1. Use --format markdown or --format html for a shareable human-readable artifact.

What the Generator Infers

  • repeated calls to a small built-in set of mutating tool names;
  • tool events that contain an error;
  • an incorrect terminal-answer signal when the final answer contains twice;
  • selected order_id, user_id, account_id, ticket_id, and file_path values from tool arguments.

These are deterministic review heuristics, not semantic evaluation of the agent, policy, or outcome.

Runtime, Data, and Network Boundary

  • The CLI reads one local JSON file and writes one local artifact.
  • It does not call a model, execute a tool, start a server, or make a network request.
  • It does not redact input. Generated cards can preserve identifiers and links, so sanitize traces before generation and review the output before sharing it.
  • Validation checks the v1 card shape and selected field types; it does not prove that the source trace or card claims are complete or correct.

Limitations

  • The generator uses fixed heuristics over recorded fields; it does not infer intent, policy compliance, or business correctness.
  • Validation proves schema shape only. It does not establish that a card is complete, accurate, or safe to share.

Compatibility

The published auraone-agent-studio-open CLI declares agent-trace-card>=0.1.1 as a runtime dependency and exposes a trace-card export path. Other trace systems can use the standalone CLI by emitting the documented local JSON shape.

Publication Status

Verified on 2026-07-13:

  • PyPI: agent-trace-card==0.1.2
  • GitHub release: v0.1.2
  • Bundled traces are synthetic tutorial fixtures, not evidence of external adoption or production performance.

Next Action

Generate a card from one sanitized failure trace, attach it to the matching incident or regression issue, and record which fields still need human context.

Downloads & popularity

Awaiting observations

No metric observations yet

Download counts and popularity will appear here as the collector receives them.

Versions 1

Collected releases
VersionPublishedDirect dependenciesOpen version
0.1.213 Jul 20260

Package advisories

0
No matching advisories collected

This describes our collected data; it is not a security assessment.

Repository

PENDING
auraoneai/agent-trace-card

Shared repository data · GitHub

License detected in repository
Not collected

Metadata history

Meaningful changes
Current collected state Current

Observed since 20 Oct, 23:34 UTC

No superseded metadata states collected. Unchanged observations do not create new revisions.

Record identity

package:8080c536c0928ff4b3b531b998867ef48114c450cd7afc05d73b93632acf6cf5