TraceVista skill
A skill and web viewer that turn Claude Code, Codex and Pi logs into a decision map of the agent run with links to originals.
- 90 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads agent session logs that may contain paths, keys and file contents
- The export keeps original records whole and must not be shared without review
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/tracevistaDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads agent session logs that may contain paths, keys and file contents
- The export keeps original records whole and must not be shared without review
This is third-party code. Review the repository files before installing.
What it does
TraceVista helps make sense of a long agent run: what it tried, why it dropped a direction and what evidence stands behind the final version. The `tracevista-skill` skill reads Claude Code, Codex and Pi logs, InferenceBench traces and SWE-rebench OpenHands trajectories, builds a state map with explanations, validates the export and saves it as `.agent-state.json`. The file opens in a web viewer at `tracevista.pages.dev`. The viewer has a decision map marking discoveries, rejected ideas and pivots, a Decisions only mode, a switch between version tree and execution order, guided playback and original records next to the explanations. `.agent-state.json` keeps the map, annotations, original records and attachments together. Conversion scripts are written in Python 3.10 and the standard library.
Who it is for. Developers and researchers who review overnight and long agent runs before accepting the result.
Good fit when
- An agent finished a multi-hour run and you need to see which approaches it dropped and why.
- A colleague must be shown the agent's course of work with evidence and not a retelling.
- Main-agent and reviewer sessions are combined into one review.
- A kernel optimization or training result is checked against the original measurement records.
Not a fit when
- Logs cannot be loaded even into a local browser: the file keeps original records with all their content.
- The agent works in a format the skill does not read: Claude Code, Codex, Pi, InferenceBench and OpenHands are supported.
- A finished report is wanted without reading the map: the viewer helps review and does not give a verdict.
Example request
Use tracevista-skill to convert this Codex log folder into a complete .agent-state.json, then open the viewer and import it.Limitations
Conversion scripts need Python 3.10 or newer. The `.agent-state.json` export includes original records and attachments, so it contains everything that was in the session, including paths and possible secrets, and the file should be reviewed before sharing. The viewer is hosted at `tracevista.pages.dev` and only the listed log formats are supported. Node explanations are written by a model, so check them against the original records shown alongside. A separate plugin handles conversion from a DeepSeek Harness session.
How to disable. Remove the skill from the agent directory or disable the plugin with the agent. Delete generated `.agent-state.json` files by hand.
FAQ
How do I install it?
Ask the agent: Install skills/tracevista-skill from https://github.com/Superone77/TraceVista. Plugin installs also exist for Claude Code, Codex, Cursor, Gemini CLI, OpenCode and Pi.
How do I invoke the skill?
In Codex with `$tracevista-skill`, in Claude Code with `/tracevista-skill` and, when installed as a plugin, with `/tracevista:tracevista-skill`.
Which logs are supported?
Claude Code, Codex and Pi sessions, InferenceBench traces and SWE-rebench OpenHands trajectories. Main-agent and reviewer sessions can be merged into one export.
Can I look without installing?
Yes, the site has example runs that open with no sign-up and no model key.
What is inside the export?
The map, annotations, original records and attachments sit in one `.agent-state.json`. It can be re-imported or sent to a colleague.
Discussion
No questions or reviews yet.
Related
Skillsopentraces records Claude Code, Codex and Pi agent work into a local log, anchors edits to commits and prepares datasets.
A local trajectory compiler that turns Claude Code and Codex session history into training data, tasks and verifiers
Thirteen skills and seven guides for Claude Code on skills, subagents, hooks and memory, checked against the July 2026 docs
Open infrastructure for agents that learn from experience: session traces, a knowledge base and skill packs via CLI, MCP and API
Local analytics for Claude Code, Codex and five more agents: tokens, cost, time and log search in one archive
A skill registry with a CLI and API: the agent describes a task and gets a recommendation with a trust score and install command
README in short
Covers the problem, example runs, installation and conversion in two prompts, viewer capabilities and supported formats.