Conversation
Add a configurable `max_retries` parameter (default: 3) that controls the number of retries the PydanticAI agent attempts on failed tool calls via `Agent(retries=...)`. The parameter flows through Config, Configuration serialization, CLI (`config set`/`config show`/`generate`), and MCP server.
perce7al
marked this pull request as ready for review
September 26, 2026 12:24
Collaborator
|
Thanks for this. It's fixed in #121 together with a few related agent-limit changes, so I'm closing this one. |
pull Bot
pushed a commit
to soitun/CodeWiki
that referenced
this pull request
Sep 29, 2026
…ons, ship updater - Package: add codewiki.src.be.updater to [tool.setuptools] packages; a non-editable install had no updater, so every --update failed on import. New test checks every package directory is listed. (FSoft-AI4Code#119) - Agent limits: pass UsageLimits(request_limit=...) to every agent run and retries=... to every Agent. pydantic-ai defaults are 50 requests per run and 1 retry per failing tool call, which complex modules and generate_sub_module_documentation hit. New settings request_limit (default 100) and agent_retries (default 3), in `codewiki config set`, `config show` and as per-run overrides on `codewiki generate`. (FSoft-AI4Code#115, FSoft-AI4Code#118) - Fallback: FallbackModel now also falls back on UnexpectedModelBehavior (a 200 response whose body does not parse), keeping ModelAPIError. (FSoft-AI4Code#117) - Overview pages: MODULE_OVERVIEW_PROMPT / REPO_OVERVIEW_PROMPT had no slot for the user's instructions. complete() takes an optional system_prompt (OpenAI- compatible, litellm, Azure, and caw via CawAgent(system_prompt=...)), and parent/repo overviews send the instructions as a system message. (FSoft-AI4Code#116) - Sub-module agents built their system prompt with a raw .format(), so with no instructions the prompt ended in the literal text "None"; they now use format_system_prompt / format_leaf_system_prompt like the top-level agents. Reported in FSoft-AI4Code#115, FSoft-AI4Code#116, FSoft-AI4Code#117, FSoft-AI4Code#118, FSoft-AI4Code#119.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add configurable
max_retriessupport for PydanticAI agents.This allows CodeWiki to control how many times an agent can retry a failed tool call before giving up.
Changes
max_retriestoConfigwith a default value of3.max_retriesthroughConfigurationserialization and backend configuration.--max-retriestoconfig set.--max-retriesas a per-run override forgenerate.config.max_retriesto all CodeWiki PydanticAI agents viaAgent(retries=...).Behavior
max_retriescontrols agent-level retries for failed tool calls. For example:allows an agent to retry a failed tool call up to three additional times before giving up.