Active Self-Improvement
WarnAudited by ClawScan on May 10, 2026.
Overview
This skill is openly designed to let the agent rewrite its own rules, memory, and behavior automatically, which is high-impact and needs careful review.
Install only if you want the agent to modify its own long-term memory and behavior. Prefer dry-run mode, keep backups of affected files, restrict editable paths, and require approval before any changes to skills, AGENTS/SOUL files, permanent memory, tickets, or cron-style reminders.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The agent could change future workflows, skills, or behavior before the user has reviewed the exact change.
The skill instructs the agent to apply some persistent changes automatically, including medium-risk changes before user approval, without defining clear limits for what counts as low or medium risk.
Low risk + reversible: Apply immediately. Log the change. Medium risk: Apply but notify user on next interaction. High risk: Write to `OUTSTANDING.md` and wait for approval.
Require explicit user approval before applying any edits to skills, protocols, behavior files, memory, tickets, or scheduled reminders.
Incorrect, malicious, or misleading content in logs or prior notes could be turned into long-term memory and reused in future tasks.
The skill reads and rewrites permanent memory based on logs and detected patterns, but does not define validation, trust boundaries, retention, or review controls for those persistent updates.
`memory/permanent/*.md` | Current knowledge state ... Stale knowledge | Update the fact | `memory/permanent/*.md`
Use dry-run by default, review all proposed memory changes, and restrict which memory files can be read or edited.
The agent may continue modifying its own operating instructions or reminders after sessions or on a schedule, rather than only when the user asks.
The skill describes autonomous self-improvement triggers outside explicit user requests, including scheduled operation, while also authorizing persistent behavior changes.
Use when the agent should get smarter without being prompted... Triggers after Recorder at end of sessions... or on a weekly cron schedule.
Disable autonomous or scheduled use unless explicitly needed, and require a visible approval step for every persistent change.