Loading ToolBench
Kernl provides 4 tools with clear, domain-specific names (kernel_workspace_*) and detailed descriptions (avg 180 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are not documented in the source, parameter constraints lack enums/ranges, and error handling guidance is absent. Tool names follow verb_noun convention well. Descriptions explain WHAT and WHEN but lack recovery guidance for failures. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels (WRITE vs READ_ONLY).
Create a new workspace owned by your office. Use to keep per-client or per-project work isolated. Set `shared: true` if you want other offices to be able to read it (they still cannot write).
Inspect a workspace's evolver state: whether it's initialised, the loaded policy, the EVOLUTION.md objectives text, the current git head, and the recent evolution-run history.
Scaffold .evolve/policy.json + EVOLUTION.md + .evolve/checks/run-all.sh inside a workspace and ensure its git repo exists. Idempotent — never overwrites existing content. Required once before run_cycle can operate on a workspace.
Run a full SEPL cycle on a workspace: snapshot baseline, dispatch the named agent with the given goal, run the policy evaluation, then either accept the candidate (eval passed) or reject + revert (eval failed). Use when supervising another agent and want its filesystem changes gated on a tests-pass / health-check signal.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., workspace_id from create response). Breaks tool chaining.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). Tools explicitly marked WRITE/READ_ONLY in metadata but not exposed to LLM via schema. Agents cannot reason about side effects.
Parameter constraints missing. 'history_limit' (number) has no min/max bounds. 'workspace_id' and 'agent_id' accept strings but no format/pattern validation. Invites invalid LLM input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Error handling lacks recovery guidance. Descriptions do not explain what to do if workspace_id is invalid, agent_id not found, or policy evaluation fails. No actionable error messages.
Idempotency not documented. 'kernel_workspace_evolution_init' claims idempotent but others (run_cycle, create) do not state whether repeated calls are safe. Agents need this to decide retry strategy.