The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands and scripts to perform maintenance and state management.
python3 scripts/memory_prep.py window is executed to compute scan windows and rotate logs.
python3 scripts/memory_prep.py stamp is executed as a final action to update watermarks.
gh pr list --state open --search "improve:" --json number,title,url is used to synchronize the memory index with GitHub activity.
git rm is invoked for log cleanup.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection due to the ingestion of untrusted data from log files.
Ingestion points: The skill scans files in memory/logs/ which are stated to contain findings from external monitors (on-chain, GitHub, papers) and other potentially untrusted summaries.
Boundary markers: The instructions lack specific delimiters or boundary markers to differentiate between the agent's memory-management instructions and the content being promoted from the logs.
Capability inventory: The skill has access to shell execution (python3, gh, git) and significant file system write/delete permissions within the repository context.
Sanitization: There is no mention of sanitization, filtering, or validation of the log content before it is promoted to the long-term MEMORY.md index or other topic files.

memory-flush