dashboard

Dashboard Skill

Analyze all Claude Code session data to produce a comprehensive usage analytics dashboard with epistemic protocol coverage metrics, friction analysis, growth timeline, and actionable improvement recommendations.

When to Use

Invoke this skill when:

• A user wants to see overall epistemic protocol usage metrics across all sessions
• Evaluating protocol coverage: which situations occurred vs. which protocols were used
• Tracking usage growth, friction patterns, and satisfaction trends over time
• After running /onboard for deeper, full-dataset analysis

Skip when:

• User wants protocol recommendations (use /onboard instead)
• Quick single-protocol question (answer directly)
• No session history exists

Workflow Overview

COLLECT → AGGREGATE → ANALYZE → PRESENT

Phase	Owner	Tool	Decision Point
1. Collect	Main	Glob	Inventory + path decision
2. Aggregate	Subagent (coverage-scanner)	Bash, Read, Grep, Glob	Batch data collection
3. Analyze	Main	—	7 computations
4. Present	Main	Write, Bash	HTML dashboard + console summary

Data Sources

Primary: Usage Data Cache

Source	Method	Extracts
~/.claude/usage-data/facets/{session_id}.json	Glob + Read	friction_counts, friction_detail, goal_categories, session_type, outcome, user_satisfaction_counts
~/.claude/usage-data/session-meta/{session_id}.json	Glob + Read	tool_counts, git_commits, git_pushes, languages, duration_minutes, start_time, first_prompt

Secondary: Session Logs

Source	Method	Extracts
~/.claude/projects/*/*.jsonl	Glob	All session JSONL files on disk
Session JSONL files	Grep command-name + "skill":"	Protocol usage history (slash commands + Skill tool invocations)

Phase Execution

Phase 1: Collect (Main)

1. Glob ~/.claude/usage-data/facets/*.json → inventory facets files
2. Glob ~/.claude/usage-data/session-meta/*.json → inventory session-meta files
3. Note JSONL glob pattern ~/.claude/projects/*/*.jsonl (scanner runs glob internally)
4. Intersect facets ∩ session-meta by filename stem (session_id)
5. Path decision: facets ∩ session-meta ≥ 10 → Path A, else Path B

If no facets or session-meta data found: report "No usage data available. Run some Claude Code sessions first, then try /dashboard again." and stop.

Phase 2: Aggregate (Subagent: coverage-scanner)

Call coverage-scanner subagent with:

• facets_dir: ~/.claude/usage-data/facets/
• session_meta_dir: ~/.claude/usage-data/session-meta/
• session_jsonl_glob: ~/.claude/projects/*/*.jsonl (scanner runs glob + grep internally, avoiding 900+ paths in prompt)
• mode: "path_a" or "path_b" based on Phase 1 decision

The subagent returns aggregated data: friction totals, outcome/satisfaction distributions, tool totals, timeline, protocol usage (slash commands + Skill tool invocations, de-duplicated), gate interaction data (per-protocol gated/relay counts), code change statistics.

Phase 3: Analyze (Main) — 9 Computations

3.1 Coverage Computation

For each protocol, determine:

• situation_occurred: whether the deficit situation was detected in session data
• situation_used: whether the protocol's slash command was actually used

Protocol	situation_occurred (Path A)	situation_occurred (Path B)
Telos	friction wrong_approach OR vague firstPrompt	vague firstPrompt
Hermeneia	friction misunderstood_request	rework (3+ edits same file)
Syneidesis	friction excessive_changes OR (wrong_approach + rework)	rework + high exploration
Aitesis	friction context_loss	N/A
Prosoche	friction wrong_file_edited	deploy/push keywords
Prothesis	—	exploration ratio ≥ 3:1
Katalepsis	—	verification firstPrompt
Epharmoge	—	N/A (conditional protocol)

situation_used detection: From Phase 2 Protocol Usage output (coverage-scanner detects via slash commands and Skill tool invocations in a single pass, de-duplicated per session). The scanner owns the mapping from skill names to protocol names — see coverage-scanner.md step 2c for the canonical mapping table.

Coverage ratio per protocol: situation_used / situation_occurred. Protocols with no detected situations = N/A.

3.2 Protocol Usage

Count protocol invocations per protocol from Phase 2 aggregated data (merged from slash commands + Skill tool). Include first usage date, detection method breakdown (command vs Skill tool), and session count if available.

3.3 Friction Mapping

Aggregate friction_counts from Phase 2. Map to protocol groups using the Tertiary Mapping Table:

Friction Key	Protocol Group
wrong_approach	Telos, Syneidesis
misunderstood_request	Hermeneia
excessive_changes	Syneidesis
context_loss	Aitesis
wrong_file_edited	Prosoche
Others (buggy_code, api_errors, etc.)	Environmental (no protocol mapping)

3.4 Growth Timeline

From Phase 2 timeline data:

• Weekly session counts
• Protocol adoption dates (first slash command usage per protocol)
• Cumulative session trend

3.5 Achievements

Compute milestones:

• Session milestones: 10, 50, 100, 250, 500 sessions
• Protocol milestones: first protocol used, 3+ protocols used, all 8 used
• Code milestones: 100 commits, 500 commits
• Streak: consecutive days with sessions (from timeline data)

3.6 Satisfaction Score

From user_satisfaction_counts (Path A only):

• Weighted average on 0-100 scale
• Split into first-half / second-half trend comparison

3.7 Quality Score

Composite score (0-100):

• Outcome (35%): success ratio from outcome distribution
• Friction rate (20%): inverse of friction per session
• Satisfaction (25%): weighted satisfaction score
• Coverage (20%): average coverage ratio across applicable protocols

3.8 Gate Efficiency

From Phase 2 gate interaction scan:

• Per-protocol: gated_interactions / total_gates ratio
• Cross-session average gate efficiency

Protocol	Avg Gate Efficiency	Relay Gates	Gated Gates	Total Sessions

Gate efficiency = 1.0 means all gates user-facing. Lower values indicate more automation via relay classification.

3.9 Relay Erosion

For protocols with 3+ session history:

• Track relay→gated transitions (erosion: gates that were auto-resolved becoming user-gated)
• Track gated→relay transitions (stabilization: gates becoming auto-resolvable with experience)
• If insufficient data (protocol used < 3 sessions): report "Insufficient data for erosion tracking."

Phase 4: Present (Main)

1. HTML Dashboard: Write to ~/.claude/.insights/dashboard.html via Write tool

   • Refer to references/html-template.md for the full HTML skeleton
   • 11 sections: Coverage, Protocol Usage, Friction→Protocol, Improvement Opportunities, Growth Timeline, Achievements, Satisfaction Trends, Quality Score, Gate Efficiency, Relay Erosion, Quick Actions
   • Path B degradation: Sections 3 (Friction), 7 (Satisfaction), 8 (Quality Score), 9 (Gate Efficiency), 10 (Relay Erosion) show "Facets data enables richer analysis" guidance. Sections 9-10 additionally show "Insufficient data for erosion tracking" when protocol session count < 3

2. Console Summary: Output key metrics:

   • Total sessions analyzed
   • Coverage summary (X/Y protocols with situations detected)
   • Top friction areas
   • Quality score (if available)
   • File path: ~/.claude/.insights/dashboard.html

3. Open in browser: Call Bash open ~/.claude/.insights/dashboard.html to launch the dashboard in the default browser

HTML Artifact Guidelines

Refer to references/html-template.md for the full HTML skeleton, CSS classes, and detailed artifact guidelines.

Rules

1. Privacy: Never transmit session data externally. All analysis runs locally.
2. Raw data in subagent: coverage-scanner subagent returns raw aggregated data only — all computation, mapping, and scoring happen in the main agent.
3. Evidence-based metrics: Every metric must be derived from actual data. No estimated or interpolated values.
4. Graceful degradation: Path B produces useful output without facets data. Sections requiring facets show clear guidance to obtain richer data.
5. Idempotent: Running /dashboard multiple times produces updated results. Previous dashboard is overwritten.
6. Subagent delegation: Phase 2 batch aggregation MUST be delegated to coverage-scanner subagent (single). Main agent handles Phases 1, 3, 4.
7. No protocol recommendations: /dashboard provides analytics only. For protocol recommendations, direct users to /onboard.