The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill contains a reference to an external documentation site (https://jeffallan.github.io/claude-skills/skills/devops/monitoring-expert/) that is blacklisted by security scanners. Blacklisted URLs pose a significant risk of distributing malware, hosting phishing content, or facilitating browser-based attacks.
[METADATA_POISONING]: The skill exhibits an author discrepancy, claiming to be authored by 'Jeffallan' (linking to their GitHub profile) while the execution environment identifies the author as 'sodiqabdulwaris'. Such inconsistencies are characteristic of impersonation attacks or supply chain compromises, especially when combined with references to flagged external resources. Additionally, the main skill file (SKILL.md) has been flagged by reputation scanners as potentially malicious.
[COMMAND_EXECUTION]: The skill provides numerous examples and instructions for executing powerful system-level tools via the command line, including clinic, py-spy, go tool pprof, and local shell scripts like ./profiler.sh. While these are standard DevOps tools, their presence in a skill linked to a malicious domain increases the risk of the agent being tricked into executing harmful commands or facilitating system exploitation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external telemetry data, which provides a surface for indirect prompt injection attacks.
Ingestion points: Data is ingested via Prometheus scrape endpoints, logging pipelines (Pino, structlog), and distributed tracing (OpenTelemetry/OTLP).
Boundary markers: The skill lacks explicit instructions or delimiters to prevent the agent from interpreting instructions embedded within ingested logs or metric metadata.
Capability inventory: The skill uses commands for installation and execution (npm, node, python, py-spy, clinic, ./profiler.sh), network communication (curl, OTLP exporters), and filesystem operations (heap snapshots).
Sanitization: There is no evidence of sanitizing or validating external input before it is processed by the monitoring persona.

monitoring-expert