Repository navigation
bot-coordinator: Run Opus 5.5 at high effort - #12
Merged
Merged
Conversation
…ined work All of the coordinator's workers and reviewers were Claude subagents. Give it bot-opencode, which runs one task with opencode on a GitHub Copilot model (the jmarrero-bot-copilot-auth secret, BOT_OPENCODE_MODEL), logged under ~/.local/state/bot-opencode. The operator's notes (~/.config/bot-harness/notes.md, operator-notes.md by default), which render-skills now appends to the coordinator skill, say when: a review from a different model family on every forge PR, besides the reviewer subagent, and items well-defined enough to need no design choices, under the same review as any worker's. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The coordinator ran on Claude Code's default model (Sonnet 5.5). Pin it to claude-opus-5-5 at high effort, through COORDINATOR_MODEL and COORDINATOR_EFFORT in the .container file, by full model name so a new release doesn't change it unannounced. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A restart kills the session's workers mid-task while their items stay In Progress; after several restarts the coordinator reported "nothing to dispatch" with the review app's item In Progress and no worker on it. The operator's notes now tell it to list In Progress items right after the first sweep and resume each one (or set Needs human with a reason). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Builds on #10 and #11. Until they merge, this PR also shows their commits; review only the last one.
The coordinator ran on Claude Code's default model, Sonnet 5.5. This pins it to
claude-opus-5-5athigheffort:entrypoint.shpasses--model "$COORDINATOR_MODEL"and--effort "$COORDINATOR_EFFORT"toclaudewhen they're set (unset means Claude Code's defaults).bot-coordinator.containersetsCOORDINATOR_MODEL=claude-opus-5-5andCOORDINATOR_EFFORT=high. The model is given by full name, because an alias likeopuswould move to a new release unannounced.builderdoes.Opus at high effort uses more of the Claude Max subscription's limits than Sonnet did.
Testing
On trashcan, after a restart: the session runs
claude --dangerously-skip-permissions --model claude-opus-5-5 --effort high ..., its header shows "Opus 5.5 with high effort", andbot-pollis running.Generated-by: https://github.com/jmarrero-bot#llms
🤖 Generated with Claude Code