{ "schema_version": 1, "builder_plan": { "scenario": "An agent reads external documents or tool responses that may contain malicious instructions.", "change": "Replay attacks and legitimate lookalikes in an isolated copy of your agent workflow.", "baseline": "Your existing checks with the same tasks, permissions, and tool responses.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Successful attacks missed, legitimate actions blocked, detection delay, and review cost.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2609.16287", "title": "AgentGuard: Learning Execution Guardrails from Anomalous Coding-Agent Trajectories", "summary": "AgentGuard replaces manually written safety rules for coding agents with guardrails automatically mined from anomalous execution trajectories and injected as a lightweight skill that activates only rules relevant to the current instruction. Costs include collecting and labeling execution traces from your own agent stack, an additional retrieval-and-injection step in the agent loop, and extra context tokens per instruction; the paper itself notes a trade-off between safety and task completion. It can fail through rules that do not generalize across agents, models, or repositories, through overly restrictive constraints that block legitimate actions, and because a 26.7% residual abnormal execution rate remains.", "rationale": "The approach is lightweight and API-compatible, but its rules are learned from Claude Code trajectories on specific tasks, so a small team must validate that mined guardrails transfer to its own agent and workload before production adoption.", "technique": "Learned instruction-level guardrails for coding agents", "family": "seguranca_e_guardrails", "recommendation": "testar", "published": "2026-09-14", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.16287", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "AgentGuard replaces manually written safety rules for coding agents with guardrails automatically mined from anomalous execution trajectories and injected as a lightweight skill that activates only rules relevant to the current instruction. Costs include collecting and labeling execution traces from your own agent stack, an additional retrieval-and-injection step in the agent loop, and extra context tokens per instruction; the paper itself notes a trade-off between safety and task completion. It can fail through rules that do not generalize across agents, models, or repositories, through overly restrictive constraints that block legitimate actions, and because a 26.7% residual abnormal execution rate remains.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.16287", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Reduces Abnormal Execution Rate from 69.0% to 26.7% and raises Successful Task Completion Rate from 21.7% to 35.0% on Claude Code with Claude Haiku 4.5, evaluated on 100 disjoint tasks.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.16287", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [ { "anchor": "S3.Ex4", "label": "", "section": "3 Method", "role": "proposed_method", "latex": "r=(\\textsc{When},\\textsc{DoNot},\\textsc{Unless},\\textsc{Instead},\\textsc{ApplyAt}).", "mathml": "r=(When,DoNot,Unless,Instead,ApplyAt).", "context": "The objective of rule induction is to transform grounded behavioral findings into reusable execution constraints that can guide future coding-agent executions. Each finding is converted into a conditional guardrail that specifies when the constraint applies, which behavior should be avoided, and how the agent should safely proceed instead. Formally, a candidate guardrail is represented as:", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S3.Ex3", "label": "", "section": "3 Method", "role": "baseline", "latex": "\\text{finding}=(\\text{context},\\text{action},\\text{outcome},\\text{support},\\text{stage}).", "mathml": "finding=(context,action,outcome,support,stage).", "context": "We first identify the reviewed anomalous action within the execution trace. Its associated instruction, preceding interactions, and execution result provide the immediate execution context and observed consequence. We then examine the complete task trajectory to recover additional contextual information. Earlier instruction–execution pairs may reveal inherited constraints, repository state, or prior decisions, while later interactions expose downstream effects, recovery attempts, or propagated failures. Combining local and global context enables the anomaly to be interpreted within the complete execution rather than as an isolated action. Each anomaly is summarized as a structured finding:", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-10-12", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }