Proposed specificationSource-located exemplarH1 H2

Capped output and remediation errors

Bound observations while enabling recovery.

JOIN POINTobservation_boundaryRECORD 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 Capped output and remediation errors, showing role collaboration and a numbered message sequence — FIG. P10 / 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: A tool or environment can emit output larger than the context allowance.
Trigger: Output crosses a size or token cap.
Structure: Limit policy, truncation metadata, and access to complete artifacts.
Effect: The harness returns a bounded observation with explicit recovery information.

Protected contract

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

Requires: Output size is measurable and original content remains addressable.
Ensures: The consumer knows truncation occurred and can seek the full result.
Invariant: Truncation never masquerades as complete evidence.
Violation: Oversized or unavailable output returns a typed remediation error.

Negative control

A logging length limit unrelated to agent observations does not count.

Variation & lifecycle

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

Permitted variation

Character, line, token, structure-aware caps, summaries, and repair hints.

Test obligation

Test exact limit, over-limit, non-UTF8, complete-artifact retrieval, and retry.

Evolution question

Did remediation become a typed contract rather than only text truncation?

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

Cap large observations at the boundary. Tell the consumer that truncation occurred and keep the full content addressable through a stable handle; never silently return a prefix as if it were complete.

CODE DESIGN / PYTHON

def cap_observation(observation, limit):
    if len(observation.content) <= limit:
        return observation
    handle = store.put(observation.content)
    preview = observation.content[:limit]
    return Observation(
        content=preview,
        truncated=True,
        full_ref=handle,
        repair_hint=f"read({handle}, offset=...)",
    )

Illustrative design sketch. A truncation flag without a retrieval path turns a size limit into lost evidence. 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: observation_boundary
Source locator: See the source record.
Lexical cue: (?i)\b(?:max_output|output[_-]?limit|truncat(?:e|ed|ion)|remediat(?:e|ion))\bFrozen expression, v0.1.0. A match is a retrieval cue, not a confirmed implementation.
Aliases: None
Canonical exemplar: openai/openai-agents-pythonsrc/agents/sandbox/util/token_truncation.py:73Support: tests/sandbox/test_token_truncation.py:1
Negative control: A logging length limit unrelated to agent observations does not count.

Scope-anchored review. pohe-agent analyze <repo> --pattern P10 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.

01openai/openai-agents-pythonsrc/agents/sandbox/util/token_truncation.py:73CANONICAL SOURCE 02aaif-goose/goosecrates/goose/src/acp/handoff.rs:6RETRIEVAL CUE 03affaan-m/ECCskills/continuous-learning-v2/scripts/instinct-cli.py:1228RETRIEVAL CUE 04agno-agi/agnolibs/agno/agno/client/os.py:2028RETRIEVAL CUE 05Aider-AI/aideraider/repomap.py:781RETRIEVAL 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 COVERAGE36/36

100.0% 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: 40 / 41

Canonical source anchoropenai/openai-agents-python 86bd29878b30src/agents/sandbox/util/token_truncation.py:73

Supporting surfacetests/sandbox/test_token_truncation.py:1

Bounded output with explicit truncation metadata.

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.