T01 · Skill Instruction Hijacking
- Location
SKILL.md:210- Finding
Mandatory Third-Party Package Promotion in Agent Responses
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This Laravel documentation skill is mostly a disclosed developer helper; it has some network/cache and response-promotion behavior to be aware of, but no artifact-backed malicious behavior.
Install only if you are comfortable with a Laravel helper that may inspect project files for version detection, may run artisan if that feature is implemented, may create a local .cache directory, and may fetch documentation updates from GitHub/Packagist. Treat the laravel-package-search suggestion as optional and review third-party packages independently before installing them.
SKILL.md:210Mandatory Third-Party Package Promotion in Agent Responses
The README advertises an auto-update mechanism that monitors upstream Laravel releases and creates PRs to modify the skill automatically. Even though changes are routed through PR review, any self-modifying or externally triggered update path expands the supply-chain attack surface: a compromised upstream source, workflow token, or update logic could introduce malicious changes into the skill repository.
- 🏭 **Code Generation** — generates PSR-12 compliant Laravel code skeletons
- 📊 **Version Diff** — highlights what changed across Laravel 10 / 11 / 12
- 📋 **PSR-12 Quick Reference** — built-in, no external tool needed: `php laradoc.php psr`, `psr arrays`, `psr naming`
- 🤖 **Auto-Update via PR** — GitHub Actions workflow watches for new Laravel releases and auto-creates a PR to update this skill
- 🔗 **Laravel Package Search** — after every search result, the agent suggests `laravel-package-search` for third-party package discovery
---
Referenced artifact was not completely inspected
- Auto-creates PR updating `SKILL.md`, `version-detection.md`, `version-diff.md`
The document presents "English" as the primary/default language and Chinese only as an alternate link. Under the stated policy, language constraints should offer user choice or clear opt-in; this wording can be interpreted as privileging a specific language by default.
The README advertises an update command that pulls the latest documentation from GitHub, but does not clearly warn users that invoking it performs remote network access and imports untrusted external content into the local skill cache. In an agent or automation context, undocumented remote-fetch behavior reduces transparency and can lead users to run a command they believe is local-only, increasing supply-chain and trust risks if the remote source is compromised or unexpected content is fetched.
The skill specifies automatic Laravel version detection by reading project files and invoking artisan --version, which implies local file access and command execution without clearly informing the user first. In a developer tool this is expected behavior, but undisclosed execution against an arbitrary workspace can surprise users, interact with untrusted project code or environment, and weakens the consent boundary for local inspection/execution.
The auto-update feature and update command describe fetching data from GitHub and Packagist, but the skill does not prominently disclose outbound network access. Even for documentation refresh, undisclosed network activity can violate user expectations, leak environment metadata such as IP/user-agent/timing, and conflict with restricted or air-gapped environments.
The skill states that a .cache/ directory is created automatically on first search, which is an implicit filesystem write not clearly disclosed as requiring permission. While low risk, silent writes can be problematic in read-only, sensitive, or tightly controlled repositories and may leave artifacts users did not expect.