The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in SKILL.md and README.md direct the agent to download and execute a shell script from the developer's domain using curl -fsSL https://tempo.xyz/install | bash.
[CREDENTIALS_UNSAFE]: The skill manages sensitive cryptographic keys and API credentials for Bridge and Stripe, storing them in the local file system at ~/.tempo/wallet/keys.toml and ~/.tempo/wallet/cards.toml. While standard for wallet software, this provides the agent with access to highly sensitive data.
[PROMPT_INJECTION]: The SKILL.md file contains instructions aimed at overriding agent behavior, specifically telling the agent to disregard summaries from web fetch tools and not to search for additional documentation.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by fetching data from arbitrary URLs and displaying the output to the agent without using boundary markers or delimiters to separate untrusted data from instructions.
Ingestion points: Untrusted data enters via crates/tempo-request/src/http/client.rs when performing HTTP requests.
Boundary markers: Absent in crates/tempo-request/src/query/output.rs; response bodies are printed directly to the output stream.
Capability inventory: The skill can execute arbitrary HTTP requests (tempo request), manage wallet keys (tempo wallet), and write data to the local file system (-o flag).
Sanitization: Control characters are stripped in crates/tempo-common/src/cli/terminal.rs for terminal safety, but the content of the data is not filtered for malicious instructions.
[DATA_EXFILTRATION]: The skill includes telemetry features that send anonymous usage data to a third-party service (PostHog), as seen in crates/tempo-common/src/analytics.rs. This can be disabled using the TEMPO_NO_TELEMETRY environment variable.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch service registry metadata and installation files from the developer's infrastructure at https://tempo.xyz and https://mpp.sh.

tempo