agent-skill-sync-manager

Warn

Audited by Gen Agent Trust Hub on Sep 23, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of shell-based commands including Git operations and a local Bash script.
  • Evidence:
  • Invocation of scripts/scan-sensitive.sh to search for patterns across the filesystem.
  • Use of git filter-repo to rewrite repository history, which is a powerful and destructive operation.
  • Use of git push --force-with-lease which can overwrite remote branch history.
  • File management commands like readlink -f, diff -rq, and git rm --cached.
  • [DATA_EXFILTRATION]: The skill's primary function is to move data from local directories to remote repositories, creating a vector for data exfiltration.
  • Evidence:
  • Instructions to push local content from ~/.agents/skills/ to a user-defined personal_remote repository.
  • Explicit handling of config.local.yml, which is described as containing "personal emails, company paths" and other sensitive metadata.
  • Mentions using SSH host aliases to authenticate against GitHub, potentially utilizing local SSH keys.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external agent skills and Git history, which creates an attack surface for instructions embedded in untrusted data.
  • Evidence:
  • Ingestion points: Processes all files within ~/.agents/skills/, the contents of config.local.yml, and the full commit history of repositories during scans.
  • Boundary markers: Absent. The skill does not implement delimiters or specific instructions to ignore embedded prompts within the files it syncs.
  • Capability inventory: Possesses significant capabilities including filesystem read/write (cp, rm), remote network transmission (git push), and repository manipulation (git filter-repo).
  • Sanitization: While it uses scan-sensitive.sh to detect data leaks, it lacks mechanisms to sanitize or ignore instructions that may be hidden in the skill files being processed.
  • [DYNAMIC_EXECUTION]: The skill utilizes external tools and scripts that generate behavior dynamically at runtime.
  • Evidence:
  • Relies on git filter-repo, a tool that dynamically rewrites commit history based on replacement rules provided in replace.txt and mailmap files.
  • The script scripts/scan-sensitive.sh dynamically constructs large regular expressions from configuration values (sensitive_patterns) and executes them using grep and git grep against the target directories and history.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 23, 2026, 02:29 AM
Security Audit — agent-trust-hub — agent-skill-sync-manager