Static source inference · medium confidence · detected: Logging
Deprecated protocol patterns detected
Summary
Statespace provides 2 tools with partial schema definitions visible in crates/statespace-tool-runtime/src/tools.rs. Tool descriptions are present but minimal (15-85 chars, below the 194-char baseline). Input schemas show type information for parameters but lack detailed descriptions for what each parameter controls. Output schemas are not documented. The 'exec' tool presents a critical security concern with irreversible shell execution capabilities and no visible safeguards. The 'curl' tool has better naming (verb-based) but both tools lack the LLM-optimized descriptions needed for confident tool selection and error recovery guidance.
Tools (2)
curlread onlysource verified55/100
Execute HTTP requests to URLs with various HTTP methods (GET, POST, PUT, PATCH, DELETE, HEAD, OPTIONS)
execirreversiblesource verified48/100
Execute arbitrary shell commands with arguments in a sandboxed environment
Tool descriptions are too brief (15-85 chars vs 194-char baseline). 'curl': 'Execute HTTP requests...' (58 chars) lacks context for when/why to use it vs other HTTP tools. 'exec': 'Execute arbitrary shell commands...' (51 chars) does not explain sandboxing details, prerequisites, or safe usage patterns.
Parameter descriptions are missing or generic. 'curl' has 'url' (bare type string, no format/constraint guidance) and 'method' (enum provided but no usage guidance). 'exec' has 'command' and 'args' with no explanation of shell context, allowed commands, escaping rules, or sandbox limits. Parameters lack the 50-150 char descriptions that guide LLM input.
Output schemas are not documented in visible source. Neither tool declares what fields the response contains, what type of data is returned, or what the agent should expect to extract. LLMs cannot plan downstream tool calls without knowing response structure.
curlexec
Recommendations
Expand 'curl' description to 150-200 chars: 'Execute HTTP requests to external URLs. Use for reading APIs, fetching data, or triggering webhooks. Includes SSRF protection and automatic header injection for auth. Does NOT support file uploads or cookie persistence. See request format details below.'
Expand 'exec' description to 150-200 chars: 'Execute shell commands in a sandboxed environment with resource limits. Use for file operations, system queries, or admin tasks. Commands run with the server's user privileges (not root). Does not support interactive input. Requires explicit allow-list in config. IRREVERSIBLE, use confirm_command first.'
Add detailed parameter descriptions for 'curl': (1) url: 'The target URL (must start with http:// or https://; no file://, ftp://, or localhost allowed). Scheme, domain, and path required (port optional, default 80/443).' (2) method: 'HTTP method, GET (fetch data, default), POST (send data), PUT (replace resource), PATCH (update fields), DELETE (remove resource), HEAD (metadata only), OPTIONS (capability check).'
Add detailed parameter descriptions for 'exec': (1) command: 'Base command name or path (e.g. grep, /usr/bin/curl). Must match allow-list in server config. Relative paths resolve from working directory. No shell metacharacters like pipes, redirects, or substitution, use args array instead.' (2) args: 'Array of command arguments passed safely (no shell expansion). Each arg is a separate parameter, do not include spaces within args; split on spaces first.'
Spec posture evidence
Inferred effective spec: <=2025-11-25.
Relies on Logging (deprecated) - log to stderr or use OpenTelemetry
'exec' tool exposes irreversible shell command execution (risk: IRREVERSIBLE) with no visible permission gates, dry-run options, or confirmation steps. No error handling guidance visible for command failures. This violates the confirmation-request and permission-gate patterns for destructive operations.
'curl' tool description claims 'various HTTP methods' but does not explain which headers are injected server-side, what authentication is handled internally, or what SSRF protections (mentioned in Cargo.toml feature 'curl') are in place. LLMs need this to avoid passing sensitive URLs or bearer tokens as parameters.
No visible error recovery guidance. Neither tool describes what happens on timeout, network failure, command crash, or permission denial. Error messages likely contain raw exit codes or stack traces rather than actionable next steps for the agent (retry? ask user? fail?).
No input validation hints in parameter descriptions. 'url' parameter has no guidance on URL format, allowed schemes, or forbidden patterns. 'command' has no hints about shell metacharacters, path resolution, or safe command prefixes. LLMs will pass unvalidated input, causing failures.
Parameter dependencies and constraints are not documented. 'curl' method enum is present but no description explains when to use each method or what 'HEAD' vs 'OPTIONS' accomplishes. 'exec' args array has no guidance on ordering, escaping, or interaction with 'command' parameter.
curlexec
Document output schemas: (1) 'curl' should return {status_code: number, headers: object, body: string, elapsed_ms: number}. (2) 'exec' should return {exit_code: number, stdout: string, stderr: string, command: string}. Add to tool descriptions and visible API docs.
Add error handling guidance to both tool descriptions: 'On failure, returns an error object with code (TIMEOUT, PERMISSION_DENIED, NETWORK_ERROR, etc.) and message. If retryable, message will suggest retry. If invalid input, message will specify the constraint violation, check and correct the parameter, then retry.'
Implement and document a dry-run mode for 'exec' (e.g. dry_run: true parameter) that validates the command and args without executing. This allows agents to confirm intent before irreversible operations.
Add SSRF protection details to 'curl' description: 'Blocks requests to private IP ranges (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16, 127.0.0.1, ::1). Blocks file:// and gopher:// schemes. Redirects followed up to 5 hops.'
Document permission requirements for 'exec': 'Requires exec:run scope. The agent must have been configured with this scope. Commands outside the server-configured allow-list will fail with PERMISSION_DENIED.'
Add timeout and resource limit parameters: 'curl' should have timeout_secs (default 30, max 300). 'exec' should have timeout_secs (default 60, max 300) and max_output_bytes (default 1MB). Document these in parameter descriptions.