session-chronicle
Warn
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile contains a 'Self-Evolving Skill' section that instructs the agent to modify the skill's own instructions ('fix this file immediately') if it encounters errors or parameter drift. This meta-instruction encourages self-modification, which can be exploited to permanently alter agent behavior. - [COMMAND_EXECUTION]: The skill utilizes several shell scripts (
s3_upload.sh,retrieve_artifact.sh,extract_session_chain.sh) to perform high-privilege system operations, including network synchronization and data compression usingaws-cliandbrotli. - [CREDENTIALS_UNSAFE]: The skill is designed to programmatically retrieve AWS credentials from 1Password using the
op readcommand and export them to the environment. While using a vault is a best practice, granting an AI agent the autonomous capability to fetch and use cloud provider secrets presents a significant security surface. - [INDIRECT_PROMPT_INJECTION]: The skill processes raw Claude Code session logs (
.jsonl) which contain untrusted data from previous conversations. If these logs contain malicious instructions, they could influence the agent during the archaeology or sanitization phases. - Ingestion points: Reads project session logs from
~/.claude/projects/inscripts/uuid_tracer.shandscripts/sanitize_sessions.py. - Boundary markers: No explicit delimiter-based isolation is used when processing the content of these logs.
- Capability inventory: The agent has access to file writing, shell execution, and network exfiltration (S3 upload).
- Sanitization: While
scripts/sanitize_sessions.pyredacts structural secrets (keys, IPs), it does not sanitize for natural language instructions that might trigger prompt injection. - [EXTERNAL_DOWNLOADS]: The
references/s3-retrieval-guide.mddocumentation suggests cloning the skill's repository fromhttps://github.com/terrylica/cc-skills.gitfor setup. This is a reference to the vendor's own repository.
Audit Metadata