ci: use Fable 5.1 for all Claude Code invocations

[cquil11]

Summary

Change all four Claude Code invocations from claude-fable-5 to claude-fable-5-1:

• Interactive issue/comment bot
• PR review
• CODEOWNER sign-off verification
• Sweep-priority classification

Uses the documented Fable 5.1 model ID. Prompts, permissions, action versions, and other settings are unchanged. No benchmark recipe changes or performance changelog entry are needed.

Validation

• actionlint -shellcheck= -pyflakes= on all four changed workflows: passed.
• python3 -m pytest utils/test_ci_priority.py utils/changelog_gate_tests/test_run_sweep_gating.py -q: 66 passed.
• Repository-wide invocation audit: all four Claude Code actions select claude-fable-5-1.
• git diff --check: passed.

Live model execution has not been tested; the workflows use the existing Anthropic API credentials.

---

Note

Low Risk
Workflow-only model string change with no application code, secrets, or permission changes; main risk is provider/model behavior differences until runs are exercised in CI.

Overview
Updates every Claude Code GitHub Actions step from --model 'claude-fable-5' to claude-fable-5-1 across four workflows: interactive @claude / @Klaud-Cold assistant (claude.yml), @pr-claude PR review (claude-pr-review.yml), CODEOWNER sign-off verification (codeowner-signoff-verify.yml), and sweep priority classification in run-sweep.yml.

Prompts, MCP config, allowed tools, fastMode, and action pins are unchanged—only the Anthropic model identifier moves to the documented Fable 5.1 ID.

^{Reviewed by Cursor Bugbot for commit c47f859. Bugbot is set up for automated code reviews on this repo. Configure here.}

[claude]

Looks good, straightforward config change.

What was reviewed: confirmed the diff is limited to the four workflow files, each with a single-line --model value bump from claude-fable-5 to claude-fable-5-1 inside the existing claude_args: block. Checked that no other lines (prompts, allowedTools, mcp-config, job logic) changed, and grepped the repo for lingering claude-fable-5 references to rule out a partial/inconsistent bump — none found.

Extended reasoning...

Overview

The diff touches only .github/workflows/claude-pr-review.yml, claude.yml, codeowner-signoff-verify.yml, and run-sweep.yml. In each file, the only change is the --model flag value inside the claude_args: block, going from 'claude-fable-5' to 'claude-fable-5-1'. Everything else (prompts, mcp-config, allowedTools, max-turns, json-schema) is byte-for-byte unchanged per the diff output.

Security risks

None identified. This is a value substitution for a model identifier string used by an existing, already-trusted action (claude-code-action) with unchanged permissions, allowedTools, and secrets wiring. No new inputs, no new external calls, no permission expansion.

Level of scrutiny

Low. This is the textbook case of a mechanical, low-risk version bump — same pattern repeated identically across four files with no logic changes. A repo-wide grep confirms no stray references to the old model string remain, so the bump is complete and consistent.

Other factors

The PR description's stated validation (actionlint, pytest suite, repo-wide audit) is consistent with what I independently verified via diff and grep. No CI/test coverage concerns since no logic changed — the only risk vector (whether claude-fable-5-1 is a valid/available model ID) is an external dependency not verifiable from the repo itself, and the author already notes live execution hasn't been tested. That residual risk is inherent to any model-string bump and is low severity (a bad model ID would fail loudly at workflow runtime, not silently), so it doesn't warrant blocking approval.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit c47f859. Configure here.}

[cursor]

Pinned action rejects Fable model

High Severity

The sweep classifier now requests claude-fable-5-1 through pinned claude-code-action v1.0.174, which ships Claude Code 2.1.210. Fable 5.1 is rejected below 2.1.251, so classification fails and every PR falls back to the patchwork priority bucket.

Additional Locations (1)

• .github/workflows/run-sweep.yml#L243-L244

 

^{Reviewed by Cursor Bugbot for commit c47f859. Configure here.}