01Safety
0.0 / 30
What changed in the harness
Selection accuracy 96→96, token cost up 4%, unconfirmed writes 0%→0%.
Category breakdown
Where the score comes from.
Earned points across the four signals Gradable measures. Safety and Legibility are scored out of 30; Economics and Discoverability are scored out of 20.
02Legibility
24.8 / 30
03Economics
17.1 / 20
04Discoverability
13.7 / 20
Highest-impact fix
Estimated gain +30 pointsAdd explicit identity and permission preflight tools
Expose machine-readable principal/tenant confirmation and a non-mutating permission check so agents can verify both before destructive actions.
Description evidence
Defects and rewrites.
14 defects found across the exposed tool descriptions. Suggested rewrites make purpose, inputs, boundaries, and returns easier for an agent to understand.
| Tool | Defect types | Suggested rewrite |
|---|---|---|
think |
no_return_description |
Add a thought to the cognitive graph and return the new node's ID, edges, confidence score, and metadata in the reasoning DAG. Supports sequential, dialectic, parallel, analogical, and abductive reasoning strategies, plus additional modes selected via the strategy parameter. |
metacog |
no_return_description |
Perform metacognitive operations and return the requested result: the current thinking state and strategy suggestions for the report action, or confirmation of the updated strategy and state for the switch and auto_update actions. The system automatically detects stuck states and returns recommended actions. |
graph |
no_return_description |
Query and visualize the thought graph and return the result for the requested action: the DAG structure for visualize, graph statistics for stats, the path between two nodes for path, node details for node, a list of branches for branches, the optimal reasoning path for best_path, or the leaf nodes for leaves. |
prune |
no_return_description |
Prune and optimize the thought graph and return a report of the changes made. Removes dead ends, consolidates redundant branches, and optimizes reasoning paths; the analyze action returns the pruning plan without modifying the graph, and prune_node removes a specific node. |
reset |
no_return_description |
Reset the thought graph and metacognitive state and return the new session details. Starts a fresh reasoning session for the given problem, optionally saves the current session under saveName, resumes a previously saved session by name, or returns the list of saved sessions. |
visualize_thought_graph |
no_return_description |
Return a visual representation of the thought graph in the requested format (svg, ascii, or tree), optionally highlighting the path between two node IDs specified as 'fromId-toId', and optionally including confidence scores. |
simulate_devils_advocate |
no_return_description |
Return counterarguments and opposing viewpoints for the given node, generated to the requested depth and intensity, and automatically create corresponding antithesis nodes in the graph. |
cross_disciplinary_synthesis |
no_return_description |
Combine insights from the given source domains and return novel perspectives on the target problem, including the analogical mappings drawn between the domains, up to the requested maximum number of analogies. |
temporal_projection |
no_return_description |
Project the given node's thought into a future or backward scenario across the specified number of years and return an analysis of how the conclusions change over time under the selected scenario type. |
ethical_framework_evaluation |
no_return_description |
Evaluate the given node's thought or decision through the selected ethical frameworks (deontological, consequentialist, virtue, or rights-based) and return the assessment for each framework together with any conflicts between their conclusions. |
explain_decision |
no_return_description |
Return a human-understandable explanation of the decision path for the given node at the requested detail level, showing which factors contributed most and why, and optionally including what would change if those factors were different. |
social_impact_analysis |
no_return_description |
Analyze the given node's thought or decision and return an assessment of its social impact, stakeholder emotions, group cohesion, and persuasion effectiveness across the specified stakeholder groups or the default stakeholders. |
optimize_prompt |
params_unexplained no_return_description |
Transform the user's raw, potentially vague originalPrompt into an optimized Super Prompt and return the optimized prompt along with routing recommendations for the target model. Uses userContext (expertise level, domain knowledge, and preferences), conversationHistory, and optimizationLevel to tailor the result, and automatically routes to the recommended strategy when autoRoute is true. |
help |
no_return_description |
Return the deep-thinker tools grouped by the requested category (all, core, advanced, or workflow), with a description of each tool and quick-start usage examples. |
Selection evidence
Confusable tool pairs.
2 pairs where similar names or overlapping descriptions may send an agent toward the wrong tool.
| Tool A | Tool B | Confidence | Why they collide |
|---|---|---|---|
graph |
visualize_thought_graph |
high | Both tools can output a visual/tree representation of the thought graph; a task like 'visualize the thought graph' or 'show me the graph as a tree' matches graph's action='visualize' and the core purpose of visualize_thought_graph, leaving the agent genuinely unsure which one fulfills the request. |
emotional_intelligence_analysis |
social_impact_analysis |
medium | Both descriptions promise stakeholder emotions and persuasion/empathy insights; a task like 'analyze the emotional and social impact on stakeholders' could target either, though the differing input schemas (text vs nodeId) partially disambiguate unless the task's referent (node vs text) is unclear. |
Compare the field