The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npx -y pocketto-pi edu --json --contract 3. The npx command downloads and executes the pocketto-pi package from the public npm registry. The -y flag suppresses the confirmation prompt, leading to the automatic execution of external, unverified code.
[EXTERNAL_DOWNLOADS]: The npx -y pocketto-pi command triggers a download of a third-party package from the npm registry. The package is not from a trusted organization or associated with the skill's author, posing a supply chain risk.
[COMMAND_EXECUTION]: The skill mandates the use of shell commands such as grep, git log, git show, and git diff to analyze code and history. While typical for debugging, these tools are executed on untrusted project data which could be crafted to exploit tool-specific vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external, untrusted data which could contain malicious instructions.
Ingestion points: CLAUDE.md, AGENTS.md, project source code, error messages, stack traces, and git diffs (as described in the Reconnaissance and Evidence Hunt phases).
Boundary markers: Absent; there are no instructions to the agent to treat this data as untrusted or to use specific delimiters.
Capability inventory: Includes tool execution capabilities such as npx, grep, and git.
Sanitization: No sanitization or validation of the input data is performed before it is processed by the agent.
[OBFUSCATION]: The skill includes a redundant binary ZIP file named bug-hunting.skill that contains a duplicate set of the skill's source files. This is an atypical distribution method that could be used to evade plain-text security scanners or provide alternate instructions.
[TIME_DELAYED_CONDITIONAL]: The execution of the npx command is gated by a specific condition: the presence of a ## Pocket Education heading in the CLAUDE.md or AGENTS.md files. This represents a conditional trigger for potentially dangerous command execution.

bug-hunting