Loading ToolBench
DevOps autopilot for Claude Code -- scan vulns, fix deps, review PRs, gen tests, track DORA metrics, multi-perspective project reviews. 21 commands, zero config.
Dobbe has 13 tools with basic Zod schemas and descriptions, but significant quality gaps limit production readiness. Tool names follow verb_noun convention (pipeline_start, config_read, cache_get), which is good. However, descriptions are often generic (e.g., 'Retrieve a cached result') and lack actionable context for LLM selection. Parameters use z.unknown() extensively (result, data, value, context), which provides zero type safety and forces LLMs to guess structure. Output schemas are completely undocumented, callers cannot know what fields to expect. Error handling returns raw error messages without recovery guidance. The pipeline_step tool accepts 'outcome' for branching but provides no enum of valid outcomes. Config and cache tools expose internal file paths and TTL mechanics that should be abstracted. No tool annotations (readOnlyHint, destructiveHint) despite clear read/write semantics.
Retrieve a cached result (e.g., previous scan results). Returns null if expired or missing.
Cache a result for future use (default TTL: 4 hours).
Read the dobbe configuration from ~/.dobbe/config.toml.
Set a configuration value in ~/.dobbe/config.toml.
Abort an in-progress pipeline session.
Finalize a completed pipeline and get the summary.
List all available pipeline commands.
Pervasive use of z.unknown() for critical parameters (result, data, value, context) eliminates type safety and forces LLMs to guess JSON structure. No validation or schema hints provided.
Output schemas completely undocumented. Callers cannot know what fields pipeline_start, pipeline_step, or cache_get return. This breaks tool chaining and forces LLMs to infer structure.
pipeline_step accepts 'outcome' parameter for branching transitions but provides no enum of valid outcomes. LLMs cannot discover valid branch keys without trial-and-error.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | <=2025-11-25 | v2 |
List all pipeline sessions (active and completed).
Start a new dobbe pipeline. Returns the first step instruction for Claude to follow.
Check the current state of a pipeline session.
Submit results for the current pipeline step and get the next instruction.
Load cross-command context (e.g., prior scan results for use in resolve).
Save cross-command context for future use.
Error handling returns raw error messages without recovery guidance. E.g., 'Error: Session not found' tells LLM nothing about next steps. Should suggest 'Call pipeline_list_sessions() to find active sessions.'
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear semantics. pipeline_abort and config_write are destructive; pipeline_list is read-only. Annotations enable safer agent planning.