The Agent Skills Directory

[COMMAND_EXECUTION]: The resume.sh script utilizes standard system utilities including find, grep, jq, ls, and sed to locate and parse local session history files. These operations are performed within the local environment and aligned with the skill's stated utility.
[DATA_EXFILTRATION]: The skill accesses and reads session transcripts stored in ~/.claude/projects/ and ~/.codex/sessions/. This data includes past conversation turns, project metadata, and file paths, which are exposed to the AI agent to facilitate session resumption.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests session data from local .jsonl files via the resume.sh script.
Boundary markers: The SKILL.md file provides explicit instructions for the agent to treat retrieved transcripts as "background context" rather than active instructions to be re-executed.
Capability inventory: The skill is granted Bash and Read tool permissions to execute its helper script and confirm the workspace state.
Sanitization: The script uses jq filters to remove specific system-generated tags (such as <system-reminder> and <local-command-caveat>) and truncates message turns exceeding 2,000 characters to mitigate context-related risks.

session-resume