Feat/nanobot integration

[chancsc]

Adds Nanobot as a 4th integration target alongside Claude Code, OpenClaw and NanoClaw.

mnemon setup --target nanobot deploys a skill file that teaches Nanobot agents all mnemon commands, workflow patterns, and guardrails

mnemon setup --eject cleanly removes the integration

Auto-detection works when the nanobot binary or ~/.nanobot/workspace/ directory is present

Recommended scope is global (--global) because Nanobot discovers skills from ~/.nanobot/workspace/ across all projects

Changes
New files:
internal/setup/nanobot.go — NanobotWriteSkill() and NanobotEject()
internal/setup/assets/nanobot/SKILL.md — skill definition with workflow, all commands, and guardrails

Modified files:
internal/setup/assets/assets.go — embed the nanobot skill asset
internal/setup/detect.go — detectNanobot() + call in DetectEnvironments()
cmd/setup.go — installNanobot(), ejectNanobot(), --target nanobot option
README.md, docs/USAGE.md, CHANGELOG.md — documentation

Integration design
Nanobot uses a skills-based model (no hooks or plugins). The agent invokes mnemon directly via Nanobot's exec() tool when the skill instructs it to. Memory is stored in the shared ~/.mnemon database, accessible across all Nanobot sessions and projects.
The SKILL.md includes guardrails: memory operations (remember/link) are delegated to spawned sub-agents to keep the main conversation clean; no secrets or passwords are stored; categories and edge types are constrained to approved values.

Test plan
make unit passes (all existing + no regressions)
make vet passes
Manual: mnemon setup --target nanobot --global --yes on a machine with Nanobot installed, verify ~/.nanobot/workspace/skills/mnemon/SKILL.md is written; mnemon setup --eject removes it cleanly

[Grivn]

Thanks, this is generally in the right direction.

A couple of small things before merge:

1. The README link currently points to dream-aim/nanobot, but the upstream project seems to be HKUDS/nanobot. Could you update it to https://github.com/HKUDS/nanobot?

2. The Nanobot skill example shows exec(command="mnemon remember ...") directly, while the guardrail says remember/link should always be delegated via spawn. Could you make those consistent? Either keeping the example recall-only or softening the guardrail to “prefer delegating remember/link when practical” would work.

Also, the PR now has a merge conflict with master, likely from the recent changelog updates. Could you rebase or merge the latest master and resolve that?

After those updates, this should be good to merge.

[Grivn]

Reviewed the updated Nanobot integration. The previous README link and skill guardrail issues are fixed, CI is passing, and local setup/eject smoke checks passed.