feat(skills): scaffold experimental skills system

[nkanf-dev]

This pull request introduces experimental support for "Agent Skills" in the mini-forensics-agent, allowing the agent to discover, activate, and utilize modular skill plugins defined in the workspace or user directories. The changes span the CLI, agent loop, prompting, and a new skills.py module, enabling dynamic skill discovery, activation, and resource access during agent runs.

The most important changes are:

Agent Skills Framework:

• Added a new skills.py module that implements skill discovery, parsing, cataloging, activation, and resource reading. Skills are defined by directories containing a SKILL.md file with YAML frontmatter and instructions. The module provides the SkillCatalog, SkillRecord, and functions for rendering and activating skills.

CLI and Workflow Integration:

• Updated the CLI (cli.py) to support new arguments: --enable-skills, --skill-dir, and --list-skills. The agent can now discover skills, list them, and enable skill-based workflows. The task argument is now optional if listing models or skills. The CLI output includes discovered skills and their roots when skills are enabled. [1] [2] [3] [4] [5]

Agent Loop and Tooling:

• Enhanced the agent loop (loop.py) to support skill activation and resource access. The loop tracks active skills, exposes skill information in prompts, and routes new tool calls (ActivateSkill, ReadSkillResource) to the appropriate handlers. Active skills are updated as the agent runs. [1] [2] [3] [4] [5] [6] [7]

Prompting and Guidance:

• Modified the prompt construction (prompting.py) to include available skills, active skill context, and new tool usage instructions when skills are enabled. The prompt guides the model on how and when to activate and use skills. [1] [2] [3] [4]

Documentation:

• Updated the README.md to document the experimental Agent Skills feature, including usage examples, skill directory precedence, skill structure, and relevant CLI commands.