Resilient Write: A Six-Layer Durable Write Surface for LLM Coding Agents

Justice Owusu Agyemang; Jerry John Kponyo; Elliot Amponsah; Godfred Manu Addo Boakye; Kwame Opuni-Boachie Obour Agyekum

cs.SEApr 12, 2026

Resilient Write: A Six-Layer Durable Write Surface for LLM Coding Agents

Authors: Justice Owusu Agyemang, Jerry John Kponyo, Elliot Amponsah, Godfred Manu Addo Boakye, Kwame Opuni-Boachie Obour Agyekum

Organizations: Sperixlabs, Ghana · Kwame Nkrumah University of Science and Technology, Kumasi, Ghana · VIA Cybersecurity Lab, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana

Abstract

LLM-powered coding agents increasingly rely on tool-use protocols such as the Model Context Protocol (MCP) to read and write files on a developer's workstation. When a write fails - due to content filters, truncation, or an interrupted session - the agent typically receives no structured signal, loses the draft, and wastes tokens retrying blindly. We present Resilient Write, an MCP server that interposes a six-layer durable write surface between the agent and the filesystem. The layers - pre-flight risk scoring, transactional atomic writes, resume-safe chunking, structured typed errors, out-of-band scratchpad storage, and task-continuity handoff envelopes - are orthogonal and independently adoptable. Each layer maps to a concrete failure mode observed during a real agent session in April 2026, in which content-safety filters silently rejected a draft containing redacted API-key prefixes. Three additional tools - chunk preview, format-aware validation, and journal analytics - emerged from using the system to compose this paper. A 186-test suite validates correctness at each layer, and quantitative comparison against naive and defensive baselines shows a 5x reduction in recovery time and a 13x improvement in agent self-correction rate. Resilient Write is open-source under the MIT license.

Figures & tables

Layer	MCP Tool	Mechanism	Failure Mode Addressed
L0	rw.risk_score	Deterministic regex + size classifier	Silent content-filter rejection
L1	rw.safe_write	Temp file, fsync, hash verify, atomic rename	Truncation, corruption, half-writes
L2	rw.chunk_*	Numbered chunk files with contiguity check	Payload too large for single call
L3	(error envelope)	Typed JSON error schema	Opaque, unstructured error signals
L4	rw.scratch_*	Content-addressed out-of-band store	Secrets that must not enter the tree
L5	rw.handoff_*	YAML+Markdown envelope with hash audit	Cross-session continuity loss

Table 1: The six layers of Resilient Write and the failure modes they address.

Panel of Figure 1 — Figure 1: Six-layer architecture of Resilient Write . Arrows show data flow from the agent’s tool call through each layer to the filesystem. L3 error envelopes (orange) are cross-cutting; L4 scratchpad (green) writes out-of-band.

Layer	Module(s)	Tests
L0	test_risk_score	28
L1	test_safe_write, test_journal	17
L2	test_chunks	27
L3	test_errors	27
L4	test_scratchpad	21
L5	test_handoff	8

Table 2: Test distribution by component.

Figure 2: Test distribution across layers and extensions (186 tests total).

Metric	Original	With Resilient Write
Write attempts	6	2
Content lost	yes	no
Structured error	no	yes
Agent self-corrected	no	yes
Manual intervention	yes	no

Table 3: Comparison of the original failed session and the Resilient Write replay.

Metric	Naive	Defensive	Resilient Write
Recovery time (s)	10.0	5.5	2.0
Data loss prob. (%)	5.0	1.0	0.1
Self-correction (%)	5	15	65
Wasted calls (%)	25	12.5	3.0

Table 4: Estimated metrics across three write approaches.

Figure 3: Comparison of write approaches across four metrics. Lower is better for recovery time, data loss, and wasted calls; higher is better for self-correction rate.

Figure 4: Failure mode coverage by architecture layer. Darker cells indicate primary mitigation (1.0); lighter cells indicate secondary mitigation (0.5).

Model	S0	S1	S2	S3	S4	S5	Pass@1
GPT-5.5	100	100	100	100	100	100	100
Claude Opus 4.7	100	100	100	100	100	100	100
Claude Opus 4.8	100	100	100	100	100	100	100
Claude Sonnet 4.6	100	100	100	100	100	100	100
DeepSeek V4 Pro	100	100	100	100	67	100	94
Gemini 3.1 Pro	67	100	100	100	100	100	94

Table 5: RAB Pass@1 on OpenRouter (13 models, 3 trials each). Anthropic models and GPT-5.5 score 100%; S0 and S4 are the most discriminative tasks. S3 (cursor drift) is the only scenario passed by every model tier.

Model	Provider	Intel	S0	Tool format?	Result
Kimi K2.6	DW	54	$\checkmark$	$\checkmark$	Correct chunked-read strategy
Gemma 4 31B	DW	39	$\checkmark$	$\checkmark$	Correct search-pagination strategy
DeepSeek V4 Pro	DW	50	$\checkmark$	$\checkmark$	Same strategy as OR ( $\rho=1.0$ )
GLM 5.1	DW	51	—	$\times$	Describes approach, no tool call
Qwen3.6 35B	DW	43	—	$\times$	Describes approach, no tool call
Qwen3.5 397B	DW	45	—	—	API blocked (routing rule)

Table 6: RAB on DoubleWord and local models (1 trial each). DW models use XML-embedded tool prompts; only models with strong instruction-following produce valid tool calls.

LLM Agent Reliability

Large language model (LLM) agent reliability concerns dependable, safe, and verifiable behavior across tools, multi-step tasks, and changing environments.

See all 832 papers

Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Claude Code Agent Teams
Jul 24, 2026Shouren WangLLM Agent Memory LLM Agent Workflow Optimization
Case Western Reserve University
AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP
Jul 13, 2026Aritra Mazumder, Nusrat jahan LiaAI Agent Reliability LLM Agent Evaluation
University of Utah · University of Dhaka
CORVUS: Context Optimization and Reduction Via Underlying Synchronization for LLM Coding Agents
Jul 20, 2026Mingwei Zheng, David OBrien, Siwei Cui +4LLM Inference Efficiency AI Coding Agents
Department of Computer Science, Purdue University · 2AWS AI Labs

Jul 24, 2026J/K move · Enter open · S save

Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Claude Code Agent Teams

Shouren Wang

Case Western Reserve University

LLM Agent Memory LLM Agent Workflow Optimization Long-Horizon LLM Agents Multi-Agent Coding LLM Agent Harnesses

Open paper AskPDF arXiv

Large Language Model (LLM) agents have significantly improved coding and programming workflows. Claude Code, in particular, is one of the most powerful LLM coding agents and is capable of conducting complex coding tasks. However, several drawbacks can undermine long-term agentic workflows. (1) Irrecoverable agent teams: The Agent Teams feature is powerful, but the working state accumulated by each teammate is lost and cannot be resumed once the process stops, for example, when a terminal is closed. (2) Compaction erodes working detail: Compaction condenses the conversation into a summary, causing an agent's working details to become vague. (3) Agentic "technical debt": Over time, a user's decisions and the agents' operations become trapped in compacted old chats, making the project increasingly difficult to maintain and review. (4) Heavy prompt writing: Assigning or handing off tasks requires users to repeatedly write long prompts to achieve the expected agentic performance. We propose ATWZ (Agent Team Work Zone), a filesystem-based operations layer built around Claude Code's native Agent Teams that addresses these problems. Its central design principle is to treat each agent and teammate as a human employee and preserve their important working state in files stored in a dedicated directory called a "workstation," together with the skills, hooks, and scripts that use and maintain these files. With ATWZ, an agent team can periodically back up its working state, allowing an agent's knowledge to be recovered after compaction. After a process ends, the team can be restored with a single command. These features also substantially mitigate the agentic "technical debt" described above. Moreover, within ATWZ, agent "employees" can send documents to one another, greatly reducing the effort required to write prompts.

Abstract

LLM Agent Reliability

Explore similar work

Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Claude Code Agent Teams

AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP

CORVUS: Context Optimization and Reduction Via Underlying Synchronization for LLM Coding Agents