Proposed specificationSource-located exemplarH2 H3

Reflection-augmented loop

Use feedback to revise attempts.

JOIN POINTfailed_attempt_feedbackRECORD 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 Reflection-augmented loop, showing role collaboration and a numbered message sequence — FIG. P20 / 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: An attempt can be evaluated before a further attempt.
Trigger: The attempt fails or receives a critique.
Structure: Attempt record, feedback producer, revision input, and retry bound.
Effect: A revised attempt uses recorded feedback while preserving attempt lineage.

Protected contract

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

Requires: Failure evidence and remaining retry budget are available.
Ensures: Feedback is attached to the attempt that produced it and is visible to revision.
Invariant: Self-critique is not represented as independent verification.
Violation: Empty, contradictory, or exhausted feedback terminates or escalates.

Negative control

A static system prompt asking the model to reflect is not enough.

Variation & lifecycle

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

Permitted variation

Self-critique, external evaluator, memory injection, and iterative refinement.

Test obligation

Test useful feedback, contradictory feedback, exhaustion, and repeated error.

Evolution question

Did reflection add independent evaluation and bounded convergence?

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

After a failed attempt, attach the failure evidence to that attempt and make it visible to the next revision. Stop when the retry budget is exhausted rather than retrying without new information.

CODE DESIGN / PYTHON

def revise(task, attempt, budget):
    if not budget.remaining:
        return Failed(task, attempt, reason="retry budget exhausted")
    feedback = evaluator.diagnose(attempt)
    next_prompt = task.prompt.with_feedback(
        attempt_id=attempt.id,
        feedback=feedback,
    )
    return attempt.run(next_prompt, budget.decrement())

Illustrative design sketch. Reflection is useful only when the feedback is linked to the attempt that produced it. 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: failed_attempt_feedback
Source locator: See the source record.
Lexical cue: (?i)\b(?:reflection|self[_-]?critique|critique|reflexion)\bFrozen expression, v0.1.0. A match is a retrieval cue, not a confirmed implementation.
Aliases: None
Canonical exemplar: OpenHands/software-agent-sdkexamples/01_standalone_sdk/31_iterative_refinement.py:370Support: tests/sdk/agent/test_iterative_refinement.py:111
Negative control: A static system prompt asking the model to reflect is not enough.

Scope-anchored review. pohe-agent analyze <repo> --pattern P20 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.

01OpenHands/software-agent-sdkexamples/01_standalone_sdk/31_iterative_refinement.py:370CANONICAL SOURCE 02affaan-m/ECCskills/council-multi-model/scripts/review-with-codex.js:46RETRIEVAL CUE 03agno-agi/agnolibs/agno/agno/db/base.py:396RETRIEVAL CUE 04Aider-AI/aideraider/io.py:1081RETRIEVAL CUE 05bytedance/deer-flowbackend/app/channels/service.py:771RETRIEVAL 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 COVERAGE29/36

80.6% 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: 32 / 41

Canonical source anchorOpenHands/software-agent-sdk 5c524ad65d81examples/01_standalone_sdk/31_iterative_refinement.py:370

Supporting surfacetests/sdk/agent/test_iterative_refinement.py:111

Attempt-feedback-refinement loop.

Exploratory retrieval cue Regex / v0.1.0

(?i)\b(?:reflection|self[_-]?critique|critique|reflexion)\b

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