The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands via Bash (specifically uv, git, and test suites) to manage project dependencies and inspect the codebase. These operations are restricted to the local environment and are necessary for the skill's primary function of dependency management.
[EXTERNAL_DOWNLOADS]: The skill retrieves package metadata and changelog content from well-known sources like PyPI (pypi.org) and GitHub. These references are used solely for informational analysis of potential breaking changes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data in the form of changelogs from the web (fetched via WebFetch and WebSearch).
Ingestion points: External package changelogs (GitHub, PyPI) and search results are processed in SKILL.md (Step 2) and references/analyse-paquet.md.
Boundary markers: The instructions use structured checklists but lack explicit 'ignore embedded instructions' delimiters for the fetched web content.
Capability inventory: The agent has access to Bash, Edit, and Agent (sub-agent delegation) capabilities.
Sanitization: There is no explicit sanitization of the external text before it is presented to the agent for analysis.
Risk Mitigation: The workflow requires a manual decision step (Step 4) and a reporting step (Step 9) that involves human oversight, which significantly mitigates the risk of the agent acting on instructions embedded in changelogs.

python-upgrade-package