Compaction Survival System
ReviewAudited by ClawScan on May 10, 2026.
Overview
This skill is transparent about preserving context, but it automatically saves broad conversation details into persistent memory files without clear limits or opt-out.
Install this only if you want the agent to keep persistent local notes about your conversations. Before using it, decide where the memory files should live, review or delete them regularly, and avoid sharing secrets unless you are comfortable with those details being written to disk.
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.
Private details shared during normal chat could be saved to local memory files and reused in later context recovery, even if the user did not intend them to become persistent state.
This directs automatic persistent logging of potentially sensitive conversation details and later exchange summaries without per-item approval, exclusions, or retention limits.
On EVERY incoming message, scan for: ... Proper nouns ... Specific values — numbers, dates, IDs, URLs, paths ... WRITE — update `SESSION-STATE.md` ... Every exchange after 60%: append human's message + your response summary
Use only in trusted workspaces, avoid sharing secrets while it is active, and add explicit controls for what may be logged, how long logs are kept, and how users can review or delete them.
The agent may create, overwrite, clear, or read memory files in the current workspace while trying to preserve context.
The skill instructs the agent to automatically mutate and read local files as part of normal operation. This is purpose-aligned, but users should notice the automatic file activity and relative paths.
Create/clear `memory/working-buffer.md` ... update `SESSION-STATE.md` ... Read `memory/working-buffer.md` ... Read today's + yesterday's `memory/YYYY-MM-DD.md`
Keep the memory files in a dedicated, clearly scoped project area and review existing files before enabling automatic create/clear behavior.
A user may interpret the privacy claim too broadly and not realize that sensitive details can still be stored locally and reused later.
The skill has no external APIs or code, but this wording could make users overlook the local retention risk from automatically writing conversation details into persistent files.
Three layers, zero dependencies, zero data leakage.
Clarify that the claim means no external network dependency, not no data retention; document local storage, review, and deletion practices.