v0.7.0

v0.7.0

Summary

AgentLane 0.7.0 adds the extensibility foundation for higher-level harness agents. This release introduces shims, skills, and first-party base tools so DefaultAgent can be used as a practical starting point for agents that shape context, expose capabilities, and operate over local project files.

Added

• Added the harness shim system for composing agent behavior around run state, prompts, tools, and lifecycle integration (04bfed0, ed6b5e7, 2eb3d26)
• Added skills support with filesystem discovery, skill parsing, activation, prompt rendering, lifecycle integration, and clinical quickstart examples that demonstrate skill-loaded context and hooks (c875a73, 9e6ceda, 310202e)
• Added first-party harness tools for reading, writing, planning, finding files, and grepping file contents, with shared prompt metadata, truncation policy, .gitignore handling, examples, and docs (a522bce, 6a41532, bd4a575, 3b5c4d4, 9f211c0, 5f54a38)

Changed

• Expanded harness documentation around shims, skills, base tools, default agents, architecture, and code style for building opinionated higher-level agents (1a9ae2a, fb751dc)
• Generalized hook integration so shims and skills can contribute hooks alongside developer-provided hooks without constraining hooks to observation-only behavior (c84a7eb, 1be5dc2)

Fixed

• Hardened native grep behavior for invalid file types, warning output, binary files, truncation, and test organization around each tool's public surface (bdc3d07, cc2cdaf)