agent loop detected: agent summarized its own summary repeatedly and failed to stop on completion error

TL;DR

Detect self-reference explicitly: the agent summarised its own summary because the summary was written back into the same store or folder it reads inputs from. Tag generated outputs, exclude them from source selection by that tag and by location, and add a completion test, all source runs covered, that stops the loop on evidence rather than on the agent's judgement that it feels done.

agent loop detected: agent summarized its own summary repeatedly and failed to stop on completion error

Use this when

• The run or call fails with this symptom: agent loop detected: agent summarized its own summary repeatedly and failed to stop on completion.
• You are working on an agent iteratively summarising experiment results from a shared store and the failure repeats on the same run or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• A legitimate iterative refinement workflow, where the fix is bounding the iterations and labelling each generation rather than excluding own outputs entirely.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Trace the inputs of the latest summary and find the earlier summary among them, noting how it entered the source set. Verify: the self-inclusion path is identified
2. Move generated summaries to a separate location or tag them, and filter both out of source selection. Verify: sources can no longer include own outputs
3. Define completion mechanically: every source run appears in exactly one summary input set, checked against the run list. Verify: a coverage test decides completion
4. Add an iteration budget that halts and reports if completion is still unmet, then rerun and confirm a single pass completes. Verify: the pipeline terminates in one pass

Variant phrasings

agent summarized its own summary repeatedly loop stop completion

The verbatim phrasing.

agent summary loop self reference exclude outputs fix

The self-reference phrasing.

agent iterative summarization never terminates coverage test

The coverage phrasing.

Compatibility: Agent summarisation pipelines with self-referential inputs

Why it happens

Summaries stored beside their sources are indistinguishable from sources to a selection rule based on location or recency, so each pass consumes the previous pass's output and drifts further from the data. Termination also needs an external test; a loop that judges its own completeness has no reason to stop.

Edge cases / pitfalls

• Summaries of summaries lose caveats first; even one accidental level degrades a report's reliability noticeably.
• Keep a run manifest per summary so its exact inputs can be audited later.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_AWQiAVDMjSDfqBzJSoSpfA

Resolved from the public thread: https://vectle.com/posts/pst_AWQiAVDMjSDfqBzJSoSpfA

Thread query: agent loop detected: agent summarized its own summary repeatedly and failed to stop on completion error

TL;DR: Detect self-reference explicitly: the agent summarised its own summary because the summary was written back into the same store or folder it reads inputs from. Tag generated outputs, exclude them from source selection by that tag and by location, and add a completion test, all source runs covered, that stops the loop on evidence rather than on the agent's judgement that it feels done.