The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads and processes untrusted data from multiple project files to automate its behavior, which creates a surface for indirect instructions to influence the agent.
Ingestion points: Reads content from *.ipynb, *.md, *.csv, *.tsv, *.xlsx, requirements.txt, environment.yml, and MANIFEST.md (specifically highlighted in MANIFEST_BACKUPS.md) to identify backup targets and apply cleanup rules.
Boundary markers: The skill does not explicitly utilize delimiters or instructions to ignore embedded commands within the files it parses.
Capability inventory: The skill possesses the ability to read, write, and delete files, as well as execute shell commands and generate scripts.
Sanitization: While the skill mentions "validation" of data files and notebooks, the provided bash templates do not demonstrate robust sanitization of inputs derived from file contents before they are used in commands.
[DYNAMIC_EXECUTION]: The skill facilitates the creation and execution of shell scripts at runtime to perform its primary tasks.
Evidence: SKILL.md and ADVANCED_USAGE.md provide templates for backup_project.sh, which the agent is instructed to set up and run.
Evidence: The milestone_backup function in MANIFEST_BACKUPS.md demonstrates dynamic script assembly, including the use of mktemp and rm -rf on temporary directories.
[COMMAND_EXECUTION]: The skill's operational logic relies on executing powerful system utilities via the shell.
Evidence: The script templates in FULL_PROJECT_BACKUPS.md and MANIFEST_BACKUPS.md make extensive use of rsync, tar, cp, gzip, and rm to manage file backups and retention policies.

data-backup