Loading ToolBench
Autonomous content generation system for note, X, and Instagram with analytics, grading, and funnel management
Content Autopilot exhibits significant definition quality gaps. Tool names lack action verbs (analytics, autopilot, dashboard, grader are nouns or gerunds, not verb_noun patterns). Descriptions are present but generic, most lack WHEN/WHY context or recovery guidance. Parameter schemas are minimal: most tools accept only 1-3 params with basic type info. No output schemas documented. Error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. The data_manager tool conflates multiple responsibilities (load/save/ensure operations). No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Composition is weak: tools operate on implicit shared state (~/.content-autopilot/) rather than explicit IDs, forcing agents to reason about file paths and state consistency.
Content analytics dashboard — deterministic calculations from content-history.json. Calculates posting frequency, funnel balance, title logic usage, category distribution, and platform activity metrics.
Content Autopilot — context analyzer and execution manifest generator. Provides context mode (returns profile and funnel balance), execute mode (returns concrete execution plan with steps), and summary mode (cumulative intelligence from history).
Content Autopilot Dashboard — generate an HTML dashboard for content performance. Creates a single-file HTML dashboard with zero external dependencies, dark theme, CSS Grid, responsive layout. All styling is inlined.
Shared JSON data manager for content-autopilot scripts. Provides consistent file I/O for profile.json, content-history.json, and active-series.json stored in ~/.content-autopilot/
Full system integrity audit — check all data files for consistency. Validates profile.json, content-history.json, active-series.json, and output directory. Reports critical, high, medium, low, and info severity issues.
Calculate funnel balance (TOFU/MOFU/BOFU) from content history. Computes stage distribution, percentages, deviations from target (50/30/20), and recommends next funnel stage based on current imbalance.
Tool names lack action verbs. 'analytics', 'autopilot', 'dashboard', 'grader' are nouns/gerunds, not verb_noun patterns (get_analytics, generate_dashboard, grade_content). LLMs cannot infer intent from noun-only names.
No output schemas documented. Tools return unstructured text or JSON, but LLMs cannot plan downstream calls without knowing field names and types. E.g., does analytics return 'posting_frequency' or 'frequency'? Does grader return 'score' or 'grade'?
data_manager conflates multiple responsibilities (load_profile, load_history, save_json, ensure_data_dir). Should split into separate tools: load_profile, load_history, save_data, ensure_directories. Single-responsibility principle violated.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Content Grader — score content quality with Japanese-aware checks. Grades hook quality, readability (paragraph length, desu/masu consistency, kanji ratio, line breaks), structure (headers, sections, conclusion), and platform-specific fitness (note character count, X tweet length, Instagram hook length and hashtags).
No error handling guidance. Tools lack descriptions of failure modes, retryability, or recovery steps. E.g., if grader fails on invalid file_path, should agent retry, ask user, or skip? No guidance provided.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). Risk levels are declared in metadata (READ_ONLY, WRITE) but not exposed in tool definitions. LLMs cannot see which tools are safe to retry or have side effects.