RFE: Add brownfield agent context discovery to apm init

[chkp-roniz]

Summary

apm init should support a preview-first discovery mode for existing projects that already contain agent context files. The goal is to inventory context used by tools such as Claude, Codex, Copilot/GitHub, Cursor, OpenCode, Windsurf, and Gemini, classify what APM can already consume, and propose an apm.yml that helps teams migrate toward reproducible agent configuration.

Proposed UX

• apm init --discover scans known project, user, and safe system-level agent context locations.
• Default mode is read-only and prints an inventory plus proposed apm.yml.
• --write writes the proposed manifest after confirmation.
• --format json|yaml|text supports automation and review workflows.

Discovery Output

Discovery should report files by:

• tool, such as Claude, Codex, Copilot/GitHub, Cursor, OpenCode, Windsurf, Gemini, APM, or shared agent skills
• scope, such as project, user, or safe system config
• type, such as instruction, agent, command, skill, context, settings, or MCP config
• importability, such as APM-native, convertible, reference-only, or ignored

Acceptance Criteria

• Plain apm init behavior remains minimal.
• Discovery reports files by tool, scope, type, and importability.
• Output avoids raw context content by default and redacts sensitive paths or credentials.
• Existing APM-native discovery is reused where possible rather than duplicated.
• Tests cover major agent folders, existing manifests, unreadable or binary files, symlinks, and redaction.

[edenfunf]

Hi, this hasn't been triaged yet. Would maintainers be open to a PR
for it? If so I'd like to pick it up and share a rough plan before
writing any code. Thanks!