Proposed specificationSource-located exemplarH2

Hybrid loop detection

Detect stalled or repetitive behavior.

JOIN POINTrepetitive_behavior_observationRECORD STATUSCandidate · v0.1.0

UML class view

The proposed contract drawn as participant roles, received calls, and numbered dependencies.

Object interactions

The same roles and calls as a class collaboration and a numbered message sequence.

Object interaction diagram for Hybrid loop detection, showing role collaboration and a numbered message sequence — FIG. P18 / OBJECT INTERACTIONS
Open full SVG Download PDF

Pointcut

Where the mechanism applies. All four dimensions belong to the proposed match, not just its vocabulary.

Context: Successive agent observations or actions can repeat without progress.
Trigger: A loop signal crosses a configured threshold or composite condition.
Structure: Repetition signals, thresholds, and an intervention interface.
Effect: The harness diagnoses a loop and stops, redirects, or escalates.

Protected contract

The obligation that clients, operators, and auditors should be able to rely on.

Requires: Action or observation history is available in comparable form.
Ensures: Intervention is attributed to explicit evidence and recorded in the trace.
Invariant: A detector cannot block all exploration solely because outputs are similar.
Violation: Detector error, disabled mode, or false positive has a recoverable path.

Negative control

A recursion depth limit alone is not semantic loop detection.

Variation & lifecycle

The hot spot is admitted only where the protected semantics survive.

Permitted variation

Action repetition, observation similarity, semantic signals, and composite thresholds.

Test obligation

Test exact repeats, productive variation, false positives, and intervention.

Evolution question

Did repetition detection move from prompt advice to executable intervention?

Prompt + code design

An illustrative prompt and a small design sketch for the proposed contract. Names and APIs are invented; this is not a repository excerpt.

PROMPT SKETCH

Watch for repetition using evidence you can name: repeated actions, near-identical observations, or a declared semantic signal. Attribute any intervention to that evidence and record it in the trace.

CODE DESIGN / PYTHON

def detect_loop(history, policy):
    window = history[-policy.window :]
    if action_repetition(window) >= policy.action_threshold:
        return Intervention("repeated action", window[-1].action)
    if observation_similarity(window) >= policy.similarity_threshold:
        return Intervention("similar observations", window[-1].observation)
    return None

Illustrative design sketch. A loop detector without a recorded reason is indistinguishable from a random nudge. This example is a teaching aid, not implementation evidence or a conformance claim.

Detection & matching

The frozen rule that retrieves this record, the canonical source exemplar, and the control that should stay unmatched.

Join point: repetitive_behavior_observation
Source locator: Barbaste et al., §6.2, p. 15 (Gemini CLI, “Hybrid Loop Detection”); Table 11 “Stuck Detection”, p. 52; source label
Lexical cue: (?i)\b(?:loop[_-]?detect(?:ion|or)?|doom[_-]?loop|repetition[_-]?detect|stuck[_-]?(?:detect|loop))\bFrozen expression, v0.1.0. A match is a retrieval cue, not a confirmed implementation.
Aliases: loopdetect, doomloop, repetitiondetect, stuckloop
Canonical exemplar: Darwin-Agent/HarnessXharnessx/processors/control/loop_detection.py:38Support: tests/unit/test_processor_triggers.py:135
Negative control: A recursion depth limit alone is not semantic loop detection.

Scope-anchored review. pohe-agent analyze <repo> --pattern P18 anchors the cue to its enclosing declaration and checks a separate operational signature for all four pointcut dimensions. Missing roles and negative-control risks remain visible; a structural match is not E2/E3. Calibration report ↗

Repository files

Up to five public files retrieved by the frozen cue and pinned to the audited commit. A match is a place to inspect, not a confirmed implementation.

01Darwin-Agent/HarnessXharnessx/processors/control/loop_detection.py:38CANONICAL SOURCE 02anomalyco/opencodepackages/core/src/v1/config/permission.ts:27RETRIEVAL CUE 03browser-use/browser-usebrowser_use/agent/service.py:440RETRIEVAL CUE 04bytedance/deer-flowbackend/packages/harness/deerflow/agents/factory.py:230RETRIEVAL CUE 05cline/clineapps/cli/src/runtime/interactive/session-config.ts:20RETRIEVAL CUE

Evidence & provenance

Each layer answers a different question. Do not substitute retrieval for review.

Source-located exemplar

A concrete source anchor is available for this proposed record. It is one exemplar, not evidence that all repositories—or every contract field—conform.

Review boundary: this candidate and its specification come from an AI-assisted study. Independent human validation remains open.

ANY-EVIDENCE FAMILY COVERAGE8/36

22.2% of repository families in the frozen panel have at least one current cue, historical cue, or canonical source layer. This is not confirmed prevalence.

Current lexical-cue repositories: 8 / 41

Canonical source anchorDarwin-Agent/HarnessX ccb411691343harnessx/processors/control/loop_detection.py:38

Supporting surfacetests/unit/test_processor_triggers.py:135

Repeated-call loop detector.

Exploratory retrieval cue Regex / v0.1.0

This expression is a discovery instrument only. A match is a retrieval cue—not a confirmed implementation, pointcut match, or contract proof.