Memory Governor

Memory governance kernel for AI agents that complements OpenClaw Dreaming with explicit correction staging, target-class routing, adapter boundaries, and saf...

This skill is internally consistent with its documentation: it provides rules, templates, and small helper scripts to check and bootstrap host-level memory adapter files. Before running anything, review the included scripts (bootstrap-generic-host.sh, check-memory-host.py, validate-memory-frontmatter.py, review-learning-candidates.py) so you understand what files they will read or create. The bootstrap script will create example files in a target directory (it documents that it won't overwrite existing files). Avoid copying secrets into the memory files it manages (references/exclusions.md explicitly forbids storing tokens/passwords). If you do not want the agent or scripts to access certain workspace files, run them in a controlled directory or inspect/limit file permissions. If you need stricter runtime guarantees, run the scripts manually rather than allowing autonomous agent invocation. Overall, the package appears to do what it claims; proceed if that governance model fits your host and you are comfortable with the scripts' filesystem access.

Type: OpenClaw Skill Name: memory-governor Version: 0.2.9 The 'memory-governor' bundle is a governance framework designed to help AI agents categorize and manage their memory layers (e.g., long-term, daily, and working buffers). The bundle contains Python validation scripts (check-memory-host.py, validate-memory-frontmatter.py) and a shell bootstrap script that operate strictly on local files to ensure workspace configuration matches the defined memory contract. Analysis of the SKILL.md and reference documentation shows no evidence of malicious prompt injection; rather, the instructions explicitly direct the agent to exclude secrets and sensitive data from memory. No network activity, data exfiltration, or unauthorized persistence mechanisms were detected.