Loading ToolBench
Local-first AI agent, MCP, provider, and dotfiles security posture auditing server. Returns redacted local security findings without mutating local config.
Nightward has 17 well-named tools with consistent verb_noun patterns (nightward_scan, nightward_findings, nightward_analysis). All tools have descriptions (avg ~80 chars, within baseline 34-392). Input schemas are present with type declarations and parameter descriptions. However, output schemas are not documented in the source, LLMs cannot predict return structure. Error handling is absent; no recovery guidance or actionable error messages visible. Most tools are READ_ONLY (safe), but write tools (nightward_action_request, nightward_action_apply_approved) lack confirmation patterns. Parameter descriptions are terse (e.g., 'optional workspace path' for workspace param in nightward_scan). No enum constraints visible for multi-value params like 'severity' in nightward_findings.
Apply an already-approved, unexpired, one-time Nightward action ticket.
Preview one bounded Nightward action before applying it.
Request local approval for one exact bounded Nightward action. This only writes Nightward approval state.
Read one Nightward action approval request status.
List bounded Nightward actions available through the shared action registry.
Run Nightward analysis with selected local or explicitly allowed online providers.
Return provider, schedule, disclosure, and settings posture.
Output schemas not documented. LLMs cannot predict return structure, field names, or types. Downstream tool chaining requires explicit response field documentation.
No error handling or recovery guidance. Tools lack actionable error messages (e.g., 'Invalid severity: must be one of critical|high|medium|low'). Agents cannot self-correct on failure.
Write tools lack confirmation/dry-run patterns. nightward_action_request and nightward_action_apply_approved modify state without preview or confirmation, risking unintended side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Return one finding by full ID or unique prefix.
Return one analysis signal by full ID or unique prefix.
Return findings with optional severity, rule, and limit filters.
Generate plan-only remediation directions for all findings, one finding, or one rule.
Run a read-only Nightward policy check.
List provider capabilities and current status.
Compare two saved report files, or the latest two saved reports when paths are omitted.
List saved scheduled report history.
List Nightward rules and remediation metadata.
Run a redacted HOME or workspace scan.
Parameter descriptions are terse and lack constraints. 'optional workspace path' does not specify format, length, or validation rules. 'filter by severity' lacks enum values.
No pagination or result limits documented. Tools like nightward_findings and nightward_rules may return unbounded lists, risking context window exhaustion.