The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local bash scripts to automate the log analysis workflow.
Evidence includes calls to scripts/find-session-files.sh, scripts/extract-subagent-calls.sh, and scripts/extract-hook-events.sh used for data collection and parsing.
[DATA_EXFILTRATION]: The skill accesses sensitive log files containing the history of past agent sessions, which represents a data exposure risk.
It searches for and reads files in directories such as ~/.claude/projects/, ~/.claude/debug/, and ~/.claude/session-env/ (File: scripts/find-session-files.sh).
These logs contain tool calls, agent transcripts, and user interactions which may contain confidential information or environmental secrets from previous sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes session logs which are external, untrusted data sources that could influence the analyzer's output or the agent's behavior.
Ingestion points: The skill reads and parses debug logs (~/.claude/debug/{sessionId}.txt) and project logs (.jsonl files) as identified in scripts/find-session-files.sh.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard potentially malicious instructions that might be embedded within the logs being analyzed.
Capability inventory: The skill uses bash for script execution, reads system logs, and performs file system checks, including verifying file deletions (File: SKILL.md, Phase 4).
Sanitization: Data extraction relies on regex-based tools like grep and sed in the provided scripts, which does not provide protection against complex injection attacks or schema confusion hidden in log data.
[PRIVILEGE_ESCALATION]: The documentation includes instructions to modify file permissions to allow script execution.
references/common-issues.md advises users to run chmod +x scripts/*.sh to resolve permission issues for the skill's own components.

session-analyzer