Smart Compact

Smart context compaction for OpenClaw agents. 4-phase progressive strategy: Scan, Extract, Check, Compact. Before running /compact, this skill scans tool out...

What to consider before installing or using this skill: - Clarify the redaction policy with the author: SKILL.md/README both say 'sensitive info will be redacted' and also list authentication tokens under 'must save' — ask which is true and request explicit examples of how tokens are redacted. - Inspect the source before installing: README points to a GitHub repo; verify that the repo and files match the published SKILL.md and that there is no hidden code that exfiltrates data. - Use 'compact check' / read-only mode first: exercise the scan and checklist phases without performing writes, to observe what the skill identifies as important. - Audit memory files: if you enable it, monitor memory/YYYY-MM-DD.md for accidental secrets and set tight filesystem permissions on the memory folder (restrict to the agent user only). - Disable automatic downstream consolidation: if you also use Memory‑Dream or other consolidation skills, ensure they are configured not to pull in these daily logs until you’re confident no secrets are stored. - Prefer testing in an isolated environment: run the skill with non-production data and simulated tool outputs to confirm behavior. - If you must allow it in production, require an explicit policy that the skill never persists raw credentials and add detection (alerting) for credential-like patterns in memory files. Given the contradictory guidance about tokens, avoid enabling persistent writes until the author provides a clear, auditable redaction approach or you instrument and verify the memory outputs yourself.

Type: OpenClaw Skill Name: pre-compact Version: 1.0.0 The smart-compact skill is a context management utility designed to prevent information loss during OpenClaw's context compaction process. It implements a four-phase strategy (Scan, Extract, Check, Compact) that instructs the agent to identify critical data such as configurations, file paths, and decisions within the conversation history and append them to local memory files (e.g., memory/YYYY-MM-DD.md). The skill emphasizes transparency, requiring user confirmation before performing the final compaction, and includes instructions to redact sensitive information, with no evidence of data exfiltration or unauthorized command execution.