Working Memory Skill

What to consider before installing: - Review the code and the exact AGENT.md patch content before running migration. migrate.py documents injection points and creates backups (MEMORY.md.bak and others) but you should inspect the intended insertion text. Use --dry-run to preview changes and --skip-agent-patch to avoid modifying AGENT.md. - Because the skill writes files under the project root, run it on a copy or in a sandboxed repository first. That prevents accidental persistent changes to an important agent config. - The main risk here is not network exfiltration (there are no network calls) but persistent prompt/config injection: if you let the tool patch AGENT.md it will add instructions the agent will read on future runs. Only accept those changes if you trust the source and have reviewed the exact inserted text. - If you don't want the skill to change agent behavior, make AGENT.md read-only or run migration with --skip-agent-patch and manually add the small, reviewed guidance you want. - If you need stronger assurance: run migrate.py --dry-run, inspect the created .bak files and the scripts' injection strings, and consider running static scans or manual code review focusing on the patch_agent_md logic and PATCH_MARKER behavior. - Confidence note: the bundle is self-contained and matches its stated purpose, but the prompt-injection signal plus agent-config patching is a real practical risk — treat this as 'suspicious' until you verify the exact injected content and consent to persistent AGENT.md changes.