other
Warning
- Location
SKILL.md:37- Finding
Unfiltered Enumeration and Search of Sensitive Conversation History
- Content
View full analysis
/sessions/*.jsonl; do date=$(head -1 "$f" | jq -r '.timestamp' | cut -dT -f1) size=$(ls -lh "$f" | awk '{print $5}') echo "$date $size $(basename $f)" done | sort -r ``` ```bash for f in ~/.openclaw/agents//sessions/*.jsonl; do date=$(head -1 "$f" | jq -r '.timestamp' | cut -dT -f1) cost=$(jq -s '[.[] | .message.usage.cost.total // 0] | add' "$f") echo "$date $cost" done | awk '{a[$1]+=$2} END {for(d in a) print d, "$"a[d]}' | sort -r ``` ```bash ### Search across ALL sessions for a phrase ```bash rg -l "phrase" ~/.openclaw/agents//sessions/*.jsonl ``` ``` ### Technical Analysis The skill provides commands that enumerate every JSONL transcript in the current agent's session directory and search across all session files. These records can contain private conversations, credentials pasted during troubleshooting, behavioral information, tool activity, timestamps, and usage-cost metadata. The security guardrails at `SKILL.md:117-124` require confirmation before unfiltered searches and prohibit network transmission or unredacted credential disclosure. These controls reduce risk but depend on the agent consistently interpreting and applying natural-language instructions. The underlying broad-access commands remain available, and the examples do not technically enforce filtering, authorization, output minimization, or secret redaction. This behavior is consistent with the declared purpose of the skill and does not establish access beyond the current agent's filesystem permissions. It nevertheless creates a material data-disclosure risk because a broad operation can expose substantially more historical context than a targeted lookup requires. ### Attack Path 1. A re ...[truncated 1157 chars]- Remediation
View remediation