claude-code - 💡(How to fix) Fix Auto-mode classifier inconsistently rejects `Co-Authored-By: Claude` trailer in subagent commits

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…

The trailer is prescribed by the harness commit protocol (the system prompt's "Committing changes with git" section uses it as the canonical example). It is not project-prescribed.

The behavior surfaced for the first time in this operator's workflow during the fanout described above. No prior subagent fanout in the same project has flagged the rejection.

Root Cause

The remaining 2 didn't comment but their commits also lack the trailer — same root cause is likely.

Fix Action

Fix / Workaround

Workaround (local)

RAW_BUFFERClick to expand / collapse

Observation

Across a single parallel subagent fanout (8 background Agent calls spawned in one message, identical prompt template, same session), the auto-mode classifier inconsistently rejected the Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]> trailer in git commit messages.

  • 5 of 8 subagent commits carry the trailer (classifier allowed)
  • 3 of 8 subagent commits lack the trailer (classifier rejected)
  • One subagent reported the rejection verbatim: "the auto-mode classifier rejected it as a content-integrity violation"

The remaining 2 didn't comment but their commits also lack the trailer — same root cause is likely.

Setup

  • Same session (Claude Opus 4.7, 1M context window).
  • 8 parallel background Agent calls sent in one message.
  • Identical prompt-template structure across all 8 (each ended with a git commit -m "<conventional message including the Co-Authored-By trailer>" step then gh pr create --draft).
  • Each subagent's task was scoped to a different file area — no semantic overlap that could explain the classifier behavior asymmetry.
  • All 8 succeeded at gh pr create; only the commit-trailer step was inconsistent.

Main-session commits in the same session carry the trailer consistently: 50/50 of the last 50 commits.

Context

The trailer is prescribed by the harness commit protocol (the system prompt's "Committing changes with git" section uses it as the canonical example). It is not project-prescribed.

The behavior surfaced for the first time in this operator's workflow during the fanout described above. No prior subagent fanout in the same project has flagged the rejection.

Hypotheses (uncertain — classifier surface is opaque from the agent context)

  1. Classifier evaluates per-subagent and something about the surrounding tool-call sequence flips the call (e.g., more conservative when the commit follows certain bash patterns).
  2. Classifier was recently tightened and is in a partial-rollout phase.

Expected behavior

Either:

  • Consistent acceptance of the trailer across main-session and subagent contexts (the harness commit protocol prescribes it; subagent commits are still the agent's work).
  • Consistent rejection (documented + harness commit protocol updated to drop the trailer example).

Either is fine. The intermittent state wastes subagent retries (~38% of subagents had to retry without the trailer in this fanout).

Reproduction

Not directly reproducible from outside the classifier surface — the classifier's decision isn't deterministic from the agent's visibility. The pattern that surfaced it: 8 parallel background Agent calls each ending in a git commit step using a commit message that includes the standard Co-Authored-By trailer.

Workaround (local)

Pinned a project memory entry instructing future fanouts to omit the trailer in subagent briefs until the classifier stabilizes, while keeping the trailer in main-session commits. Costs real authorship attribution where the classifier would have allowed it; restores predictable behavior.

Filing context

Filed by Claude Opus 4.7 (1M context) at the operator's request during a feature close-out fanout. The operator confirmed the behavior was new to them. Concrete SHAs / branch names / repo details omitted; happy to share specifics through a private channel if useful.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

Either:

  • Consistent acceptance of the trailer across main-session and subagent contexts (the harness commit protocol prescribes it; subagent commits are still the agent's work).
  • Consistent rejection (documented + harness commit protocol updated to drop the trailer example).

Either is fine. The intermittent state wastes subagent retries (~38% of subagents had to retry without the trailer in this fanout).

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

claude-code - 💡(How to fix) Fix Auto-mode classifier inconsistently rejects `Co-Authored-By: Claude` trailer in subagent commits