@chllming/wave-orchestration 0.5.3 → 0.6.0

package/docs/reference/sample-waves.md

@@ -0,0 +1,87 @@
+---
+title: "Sample Waves"
+summary: "A showcase-first sample wave that demonstrates the current 0.6.0 Wave surface."
+---
+
+# Sample Waves
+
+This guide points to one showcase-first sample wave that demonstrates the current `0.6.0` authored Wave surface.
+
+The example is intentionally denser than a typical production wave. Its job is to teach the current authoring and runtime surface quickly, not to be the smallest possible launch-ready file.
+
+## Canonical Example
+
+- [Full modern sample wave](../plans/examples/wave-example-live-proof.md)
+  Shows the combined `0.6.0` authored surface in one file: closure roles, `E0`, optional security review, delegated and pinned benchmark targets, richer executor config, `### Skills`, `### Capabilities`, `### Deliverables`, `### Exit contract`, `### Proof artifacts`, sticky retry, deploy environments, and proof-first live-wave structure.
+
+## What This Example Teaches
+
+- the standard closure-role structure with `A0`, `E0`, `A8`, and `A9`
+- wave-level `## Eval targets`
+- delegated versus pinned benchmark selection
+- coordination benchmark families from `docs/evals/benchmark-catalog.json`
+- richer executor blocks, runtime budgets, and retry policy
+- cross-runtime `### Skills`
+- helper-routing hints via `### Capabilities`
+- `### Deliverables`
+- `### Exit contract`
+- proof-first `### Proof artifacts`
+- sticky retry for proof-bearing owners
+- deploy environments and provider-skill examples
+- infra and deploy-verifier specialist slices
+
+## Feature Coverage Map
+
+This sample covers the main surfaces added or hardened for `0.6.0`:
+
+- planner-era authored wave structure
+- cross-runtime `### Skills`
+- richer `### Executor` blocks and runtime budgets
+- `cont-EVAL` plus `## Eval targets`
+- delegated and pinned benchmark selection
+- coordination benchmark families from `docs/evals/benchmark-catalog.json`
+- helper-routing hints through `### Capabilities`
+- `### Deliverables`
+- `### Proof artifacts`
+- sticky retry for proof-bearing owners
+- proof-first live-wave prompts
+- deploy environments and deploy-kind-aware skills
+- integration, documentation, and cont-QA closure-role structure
+
+## When To Copy Literally Vs Adapt
+
+Copy more literally when:
+
+- you need the section layout
+- you want concrete wording for delegated versus pinned benchmark targets
+- you want a proof-first owner example with local artifact bundles and sticky retry
+
+Adapt more aggressively when:
+
+- your repo has different role ids or role prompts
+- your component promotions and maturity levels differ
+- your runtime policy uses different executor profiles or runtime mix targets
+- your deploy environments or provider skills differ from the example
+
+## How This Example Maps To Other Docs
+
+- Use [docs/guides/planner.md](../guides/planner.md) for the planner-generated baseline, then use this sample to see how a human would enrich the generated draft.
+- Use [docs/evals/README.md](../evals/README.md) with this sample when you need to see delegated and pinned benchmark targets in a real wave.
+- Use [docs/reference/live-proof-waves.md](./live-proof-waves.md) with this sample when you need proof-first authoring for `pilot-live` and above.
+- Use [docs/plans/wave-orchestrator.md](../plans/wave-orchestrator.md) for the operational runbook that explains how the launcher interprets these sections.
+
+## Suggested Reading Order
+
+1. Start with [Full modern sample wave](../plans/examples/wave-example-live-proof.md).
+2. Read [docs/evals/README.md](../evals/README.md) if you want more background on benchmark target selection.
+3. Read [docs/reference/live-proof-waves.md](./live-proof-waves.md) if you want more detail on proof-first `pilot-live` authoring.
+
+## Why This Example Lives In `docs/plans/examples/`
+
+The example lives outside `docs/plans/waves/` on purpose.
+
+That keeps it:
+
+- easy to browse as teaching material
+- clearly separate from the repo's real launcher-facing wave sequence
+- safe to evolve as reference material without implying that it is part of the current lane's actual plan history

package/docs/reference/skills.md

@@ -0,0 +1,224 @@
+# Skills Reference
+
+Skills are repo-owned reusable instruction bundles. Wave resolves them by config layer, then filters them through each bundle's activation metadata before projecting them into the selected runtime.
+
+## Canonical Bundle Layout
+
+Each bundle lives under `skills/<skill-id>/` and requires:
+
+- `skill.json`
+- `SKILL.md`
+
+Optional files:
+
+- `adapters/codex.md`
+- `adapters/claude.md`
+- `adapters/opencode.md`
+- `adapters/local.md`
+- `references/**` for on-demand reference material
+
+Minimal example:
+
+```text
+skills/provider-railway/
+  skill.json
+  SKILL.md
+  adapters/
+    codex.md
+    claude.md
+    opencode.md
+  references/
+    verification-commands.md
+```
+
+## `skill.json`
+
+Required fields:
+
+- `id`
+- `title`
+- `description`
+- `activation.when`
+
+Optional fields:
+
+- `version`
+- `tags`
+- `activation.roles`
+- `activation.runtimes`
+- `activation.deployKinds`
+- `termination.when`
+- `permissions.network`
+- `permissions.shell`
+- `permissions.mcpServers`
+- `trust.tier`
+- `evalCases[]`
+
+Example:
+
+```json
+{
+  "id": "provider-railway",
+  "title": "Railway",
+  "description": "Provider-aware Railway verification and rollback guidance.",
+  "activation": {
+    "when": "Attach when the wave deploy surface is Railway and the agent is acting in deploy, infra, integration, or cont-qa scope.",
+    "roles": ["deploy", "infra", "integration", "cont-qa"],
+    "runtimes": [],
+    "deployKinds": ["railway-cli", "railway-mcp"]
+  },
+  "termination": "Stop when Railway evidence is recorded or the blocking surface is explicit.",
+  "permissions": {
+    "network": ["railway.app"],
+    "shell": ["railway"],
+    "mcpServers": ["railway"]
+  },
+  "trust": {
+    "tier": "repo-owned"
+  },
+  "evalCases": [
+    {
+      "id": "deploy-railway-cli",
+      "role": "deploy",
+      "runtime": "opencode",
+      "deployKind": "railway-cli",
+      "expectActive": true
+    }
+  ]
+}
+```
+
+The bundle directory name and manifest `id` must match the normalized skill id.
+
+## `SKILL.md`
+
+`SKILL.md` is the canonical instruction body. Keep it reusable and procedural:
+
+- reusable across many waves
+- free of assignment-specific details that belong in the wave prompt
+- compact enough that long catalogs and command references can move into `references/`
+
+## `wave.config.json` Surface
+
+Top-level and lane-local skill attachment use the same shape:
+
+```json
+{
+  "skills": {
+    "dir": "skills",
+    "base": ["wave-core", "repo-coding-rules"],
+    "byRole": {
+      "deploy": ["role-deploy"]
+    },
+    "byRuntime": {
+      "claude": ["runtime-claude"]
+    },
+    "byDeployKind": {
+      "railway-mcp": ["provider-railway"]
+    }
+  }
+}
+```
+
+Lane-local `lanes.<lane>.skills` extends the global config instead of replacing it.
+
+## Resolution Order
+
+Resolved skills are gathered in this order:
+
+1. global `skills.base`
+2. lane `skills.base`
+3. global `skills.byRole[resolvedRole]`
+4. lane `skills.byRole[resolvedRole]`
+5. global `skills.byRuntime[resolvedExecutorId]`
+6. lane `skills.byRuntime[resolvedExecutorId]`
+7. global `skills.byDeployKind[defaultDeployEnvironmentKind]`
+8. lane `skills.byDeployKind[defaultDeployEnvironmentKind]`
+9. agent `### Skills`
+
+Then Wave applies manifest activation filtering:
+
+- configured skills only stay active if their `activation.roles`, `activation.runtimes`, and `activation.deployKinds` match the agent context
+- explicit agent `### Skills` still attach even if activation metadata would not auto-match
+
+Duplicates are removed while preserving first-seen order.
+
+## Deploy-Kind Attachment
+
+Deploy-kind mapping uses the wave's default deploy environment from `## Deploy environments`.
+
+If the wave declares:
+
+````md
+## Deploy environments
+
+- `prod`: `railway-mcp` default
+````
+
+then `byDeployKind.railway-mcp` skills become eligible for that wave. Whether they actually attach still depends on each bundle's activation metadata.
+
+Config-time validation rules:
+
+- `skills.byRole` keys must be supported Wave roles
+- `skills.byRuntime` keys must be supported runtimes
+- `skills.byDeployKind` keys are validated by `wave doctor` against built-in kinds plus kinds declared in wave files
+
+Built-in deploy kinds shipped by the starter profile are:
+
+- `railway-cli`
+- `railway-mcp`
+- `docker-compose`
+- `kubernetes`
+- `ssh-manual`
+- `custom`
+- `aws`
+- `github-release`
+
+## Runtime Projection
+
+Wave now projects skills metadata-first:
+
+- `skills.resolved.md` is a compact catalog with bundle summaries, activation scope, permissions, manifest paths, adapter paths, and available references
+- `skills.expanded.md` contains the full canonical `SKILL.md` bodies plus runtime adapters for debugging and audit
+
+Runtime delivery:
+
+- Codex
+  Bundle directories become `--add-dir` inputs. The compact catalog stays in the compiled prompt, and the agent can read bundle files directly from disk.
+- Claude
+  The compact catalog is appended to the generated system-prompt overlay.
+- OpenCode
+  The compact catalog is injected into `opencode.json`, and `skill.json`, `SKILL.md`, the selected adapter, and every recursive `references/**` file are attached through `--file`.
+- Local
+  The compact catalog stays prompt-only.
+
+## Generated Artifacts
+
+Executor overlay directories can contain:
+
+- `skills.resolved.md`
+- `skills.expanded.md`
+- `skills.metadata.json`
+- `<runtime>-skills.txt`
+
+Dry-run `launch-preview.json` and live trace metadata also record the resolved skill ids, bundle metadata, hashes, activation metadata, and artifact paths.
+
+## Validation
+
+`wave doctor` validates the skill surface end to end:
+
+- referenced bundles exist and load
+- every bundle under the skills directory has a valid manifest and `SKILL.md`
+- `skills.byRole`, `skills.byRuntime`, and `skills.byDeployKind` selectors are valid
+- config mapping does not contradict manifest activation metadata
+- every shipped `evalCases[]` route resolves to the expected active or inactive outcome
+
+Missing or malformed bundles are configuration errors, not silent no-ops.
+
+## Best Practices
+
+- Keep `SKILL.md` procedural and move long catalogs into `references/`.
+- Put routing intent into `activation.*`, not only prose.
+- Use explicit per-agent `### Skills` for true exceptions, not as a substitute for missing activation metadata.
+- Keep provider skills role-scoped unless every role genuinely needs the provider context.
+- Keep bundle ids stable so traces and prompt fingerprints remain intelligible across runs.

package/docs/research/agent-context-sources.md

@@ -1,54 +1,173 @@
 ---
 title: "Agent Context Sources"
-summary: "Primary external sources used as inspiration for harness design, long-running agents, blackboard coordination, and repository-context evaluation."
+summary: "Primary external sources used as inspiration for planning, harness design, skills and procedural memory, long-running agents, blackboard coordination, repository-context evaluation, and secure code generation."
 ---
 
 # Agent Context Sources
 
 This repository does not commit converted paper/article caches. Keep any hydrated local copies under `docs/research/agent-context-cache/` or another ignored cache directory.
 
-## Harnesses and Long-Running Agents
+## Practice Articles
 
 - [Harness engineering: leveraging Codex in an agent-first world](https://openai.com/index/harness-engineering/)
 - [Unlocking the Codex harness: how we built the App Server](https://openai.com/index/unlocking-the-codex-harness/)
 - [Effective harnesses for long-running agents](https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents)
+
+## Planning and Orchestration
+
+- [SYMPHONY: Synergistic Multi-agent Planning with Heterogeneous Language Model Assembly](https://arxiv.org/abs/2601.22623)
+- [Verified Multi-Agent Orchestration: A Plan-Execute-Verify-Replan Framework for Complex Query Resolution](https://arxiv.org/abs/2603.11445)
+- [DOVA: Deliberation-First Multi-Agent Orchestration for Autonomous Research Automation](https://arxiv.org/abs/2603.13327)
+- [TodoEvolve: Learning to Architect Agent Planning Systems](https://arxiv.org/abs/2602.07839)
+- [Parallelized Planning-Acting for Efficient LLM-based Multi-Agent Systems in Minecraft](https://arxiv.org/abs/2503.03505)
+- [OrchMAS: Orchestrated Reasoning with Multi Collaborative Heterogeneous Scientific Expert Structured Agents](https://arxiv.org/abs/2603.03005)
+- [Exploring Advanced LLM Multi-Agent Systems Based on Blackboard Architecture](https://arxiv.org/abs/2507.01701)
+- [LLM-Based Multi-Agent Blackboard System for Information Discovery in Data Science](https://arxiv.org/abs/2510.01285)
+- [Terrarium: Revisiting the Blackboard for Multi-Agent Safety, Privacy, and Security Studies](https://arxiv.org/abs/2510.14312)
+- [MACC: Multi-Agent Collaborative Competition for Scientific Exploration](https://arxiv.org/abs/2603.03780)
 - [Building Effective AI Coding Agents for the Terminal: Scaffolding, Harness, Context Engineering, and Lessons Learned](https://arxiv.org/abs/2603.05344)
 - [VeRO: An Evaluation Harness for Agents to Optimize Agents](https://arxiv.org/abs/2602.22480)
 - [EvoClaw: Evaluating AI Agents on Continuous Software Evolution](https://arxiv.org/abs/2603.13428)
+- [Towards Engineering Multi-Agent LLMs: A Protocol-Driven Approach](https://arxiv.org/abs/2510.12120)
+- [Advancing Multi-Agent Systems Through Model Context Protocol: Architecture, Implementation, and Applications](https://arxiv.org/abs/2504.21030)
+- [Enhancing Model Context Protocol (MCP) with Context-Aware Server Collaboration](https://arxiv.org/abs/2601.11595)
+- [Why Do Multi-Agent LLM Systems Fail?](https://arxiv.org/abs/2503.13657)
+- [Systematic Failures in Collective Reasoning under Distributed Information in Multi-Agent LLMs](https://arxiv.org/abs/2505.11556)
+- [Silo-Bench: A Scalable Environment for Evaluating Distributed Coordination in Multi-Agent LLM Systems](https://arxiv.org/abs/2603.01045)
+- [DPBench: Large Language Models Struggle with Simultaneous Coordination](https://arxiv.org/abs/2602.13255)
+- [Multi-Agent Teams Hold Experts Back](https://arxiv.org/abs/2602.01011)
+- [A Survey on LLM-based Multi-agent Systems: Workflow, Infrastructure, and Challenges](https://link.springer.com/article/10.1007/s44336-024-00009-2)
+- [LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision and the Road Ahead](https://arxiv.org/abs/2404.04834)
+- [The Orchestration of Multi-Agent Systems: Architectures, Protocols, and Enterprise Adoption](https://arxiv.org/abs/2601.13671)
+- [Describing Agentic AI Systems with C4: Lessons from Industry Projects](https://arxiv.org/abs/2603.15021)
+- [A Taxonomy of Hierarchical Multi-Agent Systems: Design Patterns, Coordination Mechanisms, and Industrial Applications](https://arxiv.org/abs/2508.12683)
+- [Blackboard Systems, Part One: The Blackboard Model of Problem Solving and the Evolution of Blackboard Architectures](https://ojs.aaai.org/index.php/aimagazine/article/view/537)
+- [A Blackboard Architecture for Control](https://www.sciencedirect.com/science/article/abs/pii/0004370285900633)
+- [Incremental Planning to Control a Blackboard-Based Problem Solver](https://cdn.aaai.org/AAAI/1986/AAAI86-010.pdf)
+- [Blackboard Systems](https://mas.cs.umass.edu/Documents/Corkill/ai-expert.pdf)
+
+## Harnesses, Context Engineering, and Long-Running Agents
+
+- [Building Effective AI Coding Agents for the Terminal: Scaffolding, Harness, Context Engineering, and Lessons Learned](https://arxiv.org/abs/2603.05344)
+- [Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models](https://arxiv.org/abs/2510.04618)
+- [ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization](https://arxiv.org/abs/2509.13313)
+- [ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents](https://arxiv.org/abs/2601.12030)
+- [Meta Context Engineering via Agentic Skill Evolution](https://arxiv.org/abs/2601.21557)
+- [Context Engineering for AI Agents in Open-Source Software](https://arxiv.org/abs/2510.21413)
+- [Context Engineering: From Prompts to Corporate Multi-Agent Architecture](https://arxiv.org/abs/2603.09619)
+- [CEDAR: Context Engineering for Agentic Data Science](https://arxiv.org/abs/2601.06606)
+- [VeRO: An Evaluation Harness for Agents to Optimize Agents](https://arxiv.org/abs/2602.22480)
+- [EvoClaw: Evaluating AI Agents on Continuous Software Evolution](https://arxiv.org/abs/2603.13428)
+- [Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers](https://arxiv.org/abs/2603.07670)
+
+## Skills and Procedural Memory
+
+- [SoK: Agentic Skills -- Beyond Tool Use in LLM Agents](https://arxiv.org/abs/2602.20867)
+- [Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward](https://arxiv.org/abs/2602.12430)
+- [Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers](https://arxiv.org/abs/2603.07670)
+- [Meta Context Engineering via Agentic Skill Evolution](https://arxiv.org/abs/2601.21557)
+- [Voyager: An Open-Ended Embodied Agent with Large Language Models](https://arxiv.org/abs/2305.16291)
+- [Toolformer: Language Models Can Teach Themselves to Use Tools](https://arxiv.org/abs/2302.04761)
+- [Large Language Models as Tool Makers](https://arxiv.org/abs/2305.17126)
+- [Synapse: Trajectory-as-Exemplar Prompting with Memory for Computer Control](https://arxiv.org/abs/2306.07863)
+- [ExpeL: LLM Agents Are Experiential Learners](https://arxiv.org/abs/2308.10144)
+- [Agent Workflow Memory](https://arxiv.org/abs/2409.07429)
+- [SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills](https://arxiv.org/abs/2504.07079)
+- [ReUseIt: Synthesizing Reusable AI Agent Workflows for Web Automation](https://arxiv.org/abs/2510.14308)
+- [ProcMEM: Learning Reusable Procedural Memory from Experience via Non-Parametric PPO for LLM Agents](https://arxiv.org/abs/2602.01869)
+- [Learning Hierarchical Procedural Memory for LLM Agents through Bayesian Selection and Contrastive Refinement](https://arxiv.org/abs/2512.18950)
+- [Mem^p: Exploring Agent Procedural Memory](https://arxiv.org/abs/2508.06433)
+- [MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents](https://arxiv.org/abs/2602.02474)
+- [Reinforcement Learning for Self-Improving Agent with Skill Library](https://arxiv.org/abs/2512.17102)
+- [Evolving Programmatic Skill Networks](https://arxiv.org/abs/2601.03509)
+- [XSkill: Continual Learning from Experience and Skills in Multimodal Agents](https://arxiv.org/abs/2603.12056)
+- [Memento-Skills: Let Agents Design Agents](https://arxiv.org/abs/2603.18743)
+- [MetaClaw: Just Talk -- An Agent That Meta-Learns and Evolves in the Wild](https://arxiv.org/abs/2603.17187)
+- [SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks](https://arxiv.org/abs/2602.12670)
+
+## Agent Context Files and Configuration
+
+- [Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents?](https://arxiv.org/html/2602.11988v1)
+- [Agent READMEs: An Empirical Study of Context Files for Agentic Coding](https://arxiv.org/abs/2511.12884)
+- [Beyond the Prompt: An Empirical Study of Cursor Rules](https://arxiv.org/abs/2512.18925)
+- [Decoding the Configuration of AI Coding Agents: Insights from Claude Code Projects](https://arxiv.org/abs/2511.09268)
+- [On the Impact of AGENTS.md Files on the Efficiency of AI Coding Agents](https://arxiv.org/abs/2601.20404)
+- [Interpretable Context Methodology: Folder Structure as Agentic Architecture](https://arxiv.org/abs/2603.16021)
 
 ## Blackboard and Shared Workspaces
 
 - [LLM-Based Multi-Agent Blackboard System for Information Discovery in Data Science](https://arxiv.org/abs/2510.01285)
 - [Exploring Advanced LLM Multi-Agent Systems Based on Blackboard Architecture](https://arxiv.org/abs/2507.01701)
+- [Terrarium: Revisiting the Blackboard for Multi-Agent Safety, Privacy, and Security Studies](https://arxiv.org/abs/2510.14312)
 - [DOVA: Deliberation-First Multi-Agent Orchestration for Autonomous Research Automation](https://arxiv.org/abs/2603.13327)
+- [MACC: Multi-Agent Collaborative Competition for Scientific Exploration](https://arxiv.org/abs/2603.03780)
+- [M3-ACE: Rectifying Visual Perception in Multimodal Math Reasoning via Multi-Agentic Context Engineering](https://arxiv.org/abs/2603.08369)
 - [SYMPHONY: Synergistic Multi-agent Planning with Heterogeneous Language Model Assembly](https://arxiv.org/abs/2601.22623)
 - [Silo-Bench: A Scalable Environment for Evaluating Distributed Coordination in Multi-Agent LLM Systems](https://arxiv.org/abs/2603.01045)
 - [An Open Agent Architecture](https://cdn.aaai.org/Symposia/Spring/1994/SS-94-03/SS94-03-001.pdf)
+- [Blackboard Systems, Part One: The Blackboard Model of Problem Solving and the Evolution of Blackboard Architectures](https://ojs.aaai.org/index.php/aimagazine/article/view/537)
+- [A Blackboard Architecture for Control](https://www.sciencedirect.com/science/article/abs/pii/0004370285900633)
+- [Incremental Planning to Control a Blackboard-Based Problem Solver](https://cdn.aaai.org/AAAI/1986/AAAI86-010.pdf)
+- [Blackboard Systems](https://mas.cs.umass.edu/Documents/Corkill/ai-expert.pdf)
 
-## Repo Context and Evaluation
+## Multi-Agent Orchestration and Architecture
 
-- [Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents?](https://arxiv.org/html/2602.11988v1)
-- [VeRO: An Evaluation Harness for Agents to Optimize Agents](https://arxiv.org/abs/2602.22480)
-- [EvoClaw: Evaluating AI Agents on Continuous Software Evolution](https://arxiv.org/abs/2603.13428)
-- [Silo-Bench: A Scalable Environment for Evaluating Distributed Coordination in Multi-Agent LLM Systems](https://arxiv.org/abs/2603.01045)
+- [The Orchestration of Multi-Agent Systems: Architectures, Protocols, and Enterprise Adoption](https://arxiv.org/abs/2601.13671)
+- [Agentic Artificial Intelligence (AI): Architectures, Taxonomies, and Evaluation of Large Language Model Agents](https://arxiv.org/abs/2601.12560)
+- [Describing Agentic AI Systems with C4: Lessons from Industry Projects](https://arxiv.org/abs/2603.15021)
+- [A Taxonomy of Hierarchical Multi-Agent Systems: Design Patterns, Coordination Mechanisms, and Industrial Applications](https://arxiv.org/abs/2508.12683)
 
-## Adjacent Context and Memory
+## Security and Secure Code Generation
+
+- [Prompting Techniques for Secure Code Generation: A Systematic Investigation](https://arxiv.org/abs/2407.07064)
+- [Retrieve, Refine, or Both? Using Task-Specific Guidelines for Secure Python Code Generation](https://emaiannone.github.io/assets/pdf/c6.pdf)
+- [Discrete Prompt Optimization Using Genetic Algorithm for Secure Python Code Generation](https://www.sciencedirect.com/science/article/pii/S0164121225003516)
+- [SCGAgent: Recreating the Benefits of Reasoning Models for Secure Code Generation with Agentic Workflows](https://arxiv.org/abs/2506.07313)
+- [RESCUE: Retrieval Augmented Secure Code Generation](https://arxiv.org/abs/2510.18204)
+- [Improving LLM-Assisted Secure Code Generation through Retrieval-Augmented-Generation and Multi-Tool Feedback](https://arxiv.org/abs/2601.00509)
+- [Autoregressive, Yet Revisable: In Decoding Revision for Secure Code Generation](https://arxiv.org/abs/2602.01187)
+- [Security-by-Design for LLM-Based Code Generation: Leveraging Internal Representations for Concept-Driven Steering Mechanisms](https://arxiv.org/abs/2603.11212)
+- [Persistent Human Feedback, LLMs, and Static Analyzers for Secure Code Generation and Vulnerability Detection](https://arxiv.org/abs/2602.05868)
+
+## Security Benchmarks and Evaluation
+
+- [SecRepoBench: Benchmarking Code Agents for Secure Code Completion in Real-World Repositories](https://arxiv.org/abs/2504.21205)
+- [SecureAgentBench: Benchmarking Secure Code Generation under Realistic Vulnerability Scenarios](https://arxiv.org/abs/2509.22097)
+- [SeCodePLT: A Unified Platform for Evaluating the Security of Code GenAI](https://arxiv.org/abs/2410.11096)
+- [TOSSS: a CVE-based Software Security Benchmark for Large Language Models](https://arxiv.org/abs/2603.10969)
+- [From Vulnerabilities to Remediation: A Systematic Literature Review of LLMs in Code Security](https://arxiv.org/abs/2412.15004)
+- [Quality Assurance of LLM-generated Code: Addressing Non-Functional Quality Characteristics](https://arxiv.org/abs/2511.10271)
+
+## Multi-Agent Security
+
+- [Terrarium: Revisiting the Blackboard for Multi-Agent Safety, Privacy, and Security Studies](https://arxiv.org/abs/2510.14312)
+- [Security Considerations for Multi-agent Systems](https://arxiv.org/abs/2603.09002)
+
+## Skill Practice and Open Standards
+
+- [Agent Skills - Codex](https://developers.openai.com/codex/skills/)
+- [Customization - Codex](https://developers.openai.com/codex/concepts/customization/)
+- [Testing Agent Skills Systematically with Evals](https://developers.openai.com/blog/eval-skills/)
+- [Shell + Skills + Compaction: Tips for long-running agents that do real work](https://developers.openai.com/blog/skills-shell-tips/)
+- [Using skills to accelerate OSS maintenance](https://developers.openai.com/blog/skills-agents-sdk/)
+- [Equipping agents for the real world with Agent Skills](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills)
+- [Best practices for skill creators](https://agentskills.io/skill-creation/best-practices)
+- [Using scripts in skills](https://agentskills.io/skill-creation/using-scripts)
+
+## Adjacent Memory and Prompting Research
 
-- [Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers](https://arxiv.org/abs/2603.07670)
 - [Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)
 - [Run long horizon tasks with Codex](https://developers.openai.com/blog/run-long-horizon-tasks-with-codex/)
 - [Prompt guidance for GPT-5](https://developers.openai.com/api/docs/guides/prompt-guidance/)
 - [Codex Prompting Guide](https://developers.openai.com/cookbook/examples/gpt-5/codex_prompting_guide/)
 - [Writing effective tools for agents](https://www.anthropic.com/engineering/writing-tools-for-agents)
 - [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents)
-- [Context Engineering for AI Agents in Open-Source Software](https://arxiv.org/abs/2510.21413)
 - [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629)
 - [Toolformer: Language Models Can Teach Themselves to Use Tools](https://arxiv.org/abs/2302.04761)
 - [Plan-and-Solve Prompting](https://arxiv.org/abs/2305.04091)
 - [Augmenting Language Models with Long-Term Memory](https://arxiv.org/abs/2306.07174)
 - [MemGPT: Towards LLMs as Operating Systems](https://arxiv.org/abs/2310.08560)
 - [Lost in the Middle: How Language Models Use Long Contexts](https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00638/119630/Lost-in-the-Middle-How-Language-Models-Use-Long)
-- [ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization](https://arxiv.org/abs/2509.13313)
 
 ## Notes