This capability is currently in public preview. It is available to every customer and Fiddler responds promptly to issues, but it is not yet covered by an SLA and its APIs may still change.
fiddler-claude-code-plugin, published from the open-source
fiddler-labs/fiddler-coding-agent-plugin repository.
To install or move to a newer version, see the
installation and upgrade guide.
Versions earlier than 0.6.0 were private-preview builds and are not listed here.
0.6
- Changes
- No
ghCLI required: The plugin now downloads its binary from the public GitHub Release over HTTPS withcurl, falling back towgetifcurlis not installed. You no longer need theghCLI or GitHub authentication on machines that run the plugin. - Checksum verification unchanged: The plugin still installs a downloaded binary only after its SHA-256 checksum matches the release’s
checksums.txt, which requiressha256sumorshasum. If verification fails, the plugin skips tracing rather than run an unverified binary. - Machines that run the plugin need outbound HTTPS access to
github.comand to the release-asset host that GitHub redirects downloads to (currentlyrelease-assets.githubusercontent.com).
- No
-
Initial Public Release
- First release from the public, Apache 2.0-licensed
fiddler-labs/fiddler-coding-agent-pluginrepository. Add the marketplace and install the plugin as described in the installation guide. - This version still downloads its binary with the
ghCLI. Upgrade to 0.6.1 to remove that requirement.
- First release from the public, Apache 2.0-licensed
-
Bug Fixes
- Final LLM span no longer dropped at turn end: Claude Code can finish a turn before it writes the turn’s final assistant message to the session transcript. Previously, when the plugin read the transcript first, the turn’s final LLM span was missing, or was present without its response text. The plugin now waits up to 1 second for the final message, only when this race occurs (typically taking about 100 ms). If only the response text is missing, the plugin fills it into the existing span. If the whole message never arrives, the plugin builds the final LLM span from the response that Claude Code reports at turn end; that span has no model or token usage and carries
fiddler.coding_agent.llm_call.synthesized=true.
- Final LLM span no longer dropped at turn end: Claude Code can finish a turn before it writes the turn’s final assistant message to the session transcript. Previously, when the plugin read the transcript first, the turn’s final LLM span was missing, or was present without its response text. The plugin now waits up to 1 second for the final message, only when this race occurs (typically taking about 100 ms). If only the response text is missing, the plugin fills it into the existing span. If the whole message never arrives, the plugin builds the final LLM span from the response that Claude Code reports at turn end; that span has no model or token usage and carries
-
Removals
FIDDLER_OMIT_IOsetting removed: This environment variable was never enforced; prompt, response, and tool input/output content was always captured regardless of its value. It has been removed so it no longer suggests that content is redacted. See Data Capture and Privacy.
-
Upgrading from an Earlier Build
- Refresh the marketplace and install the latest version, then run
/reload-pluginsif Claude Code prompts you. Your existing configuration is kept. See Upgrading.
- Refresh the marketplace and install the latest version, then run