Skylv Capability Growth

Tracks and grows agent capabilities over time

Security Scan

The skill's behavior (local log parsing and reporting) matches its description, but the provided JavaScript file is truncated in the package listing so I cannot verify there are no hidden network calls or exfiltration; also the tool will recursively read any files under the supplied logs directory which could expose sensitive data if misused.

Purpose & Capability

Name and description match the actual behavior: the SKILL.md and growth_engine.js describe parsing local session logs and producing capability-growth reports. No unrelated credentials, binaries, or install steps are requested.

Instruction Scope

Runtime instructions are scoped to running node growth_engine.js against a logs directory. The code reads and scores files, extracts wins/failures, and generates reports. However, the engine recursively traverses the provided directory and will parse any .md/.txt/.log files there, so if the user points it at a directory containing secrets or unrelated system files it will read them. Also the growth_engine.js content in the listing is truncated, so I cannot confirm there are no additional behaviors later in the file (e.g., network calls, child_process exec, telemetry).

Install Mechanism

No install spec; it's an instruction-only skill with an included Node.js script. Nothing is downloaded or written by an installer step.

Credentials

The skill requests no environment variables, credentials, or config paths. The code filters some token-like strings from reported wins (e.g., ghp_), which is a sign it expects tokens may appear in logs, but it does not declare any secret access.

Persistence & Privilege

always is false and the skill is user-invocable. It can be invoked autonomously (platform default), but it does not request permanent system presence or modify other skills. Autonomous invocation combined with file-reading capability increases blast radius if misused—note this is platform-default behavior.

What to consider before installing

This skill appears to do what it says: local log analysis and report generation. Before installing or running it: - Review the full growth_engine.js source (the provided file listing was truncated). Specifically check for any network requests, child_process.exec/spawn usage, or code that writes or uploads data. - Run it against a safe copy of your logs (or a sanitized subset), not your entire home directory, because it will recursively read .md/.txt/.log files and may surface sensitive content in reports. - If you must run it on real memory/workspace data, run it in a restricted environment (container or VM) with no network access so any unexpected exfiltration would be prevented. - Look for handling of secrets in your logs: the script filters a few token patterns but may not catch all API keys—sanitize or redact before analysis. If you can provide the remaining portion of growth_engine.js or confirm there are no outgoing network calls, I can raise the confidence and potentially change the verdict to benign.

Track your AI agent's capability growth over time. Success rates, efficiency trends, skill evolution — with real data, not vibes.

Skill Metadata

• Slug: skylv-capability-growth
• Version: 1.0.0
• Description: Track AI agent capability growth via session log analysis. Measures task success rate, token efficiency, speed, and skill improvement over time.
• Category: agent
• Trigger Keywords: growth, capability, improvement, evolution, performance, metrics, success rate, efficiency, progress

---

What It Does

Analyzes a directory of session logs (.md daily logs, conversation exports, or any text files) and produces a capability growth report:

node growth_engine.js analyze <logsDir> [--period days]
node growth_engine.js trend <logsDir> [--metric success|token|speed|time]
node growth_engine.js compare <logsDir> [--period1 N] [--period2 M]
node growth_engine.js report <logsDir> [--format markdown|json]

Example Output

## Capability Growth Report
Period: 2026-03-20 → 2026-04-17 (28 days)

### 📈 Task Success Rate
Week 1:  ████░░░░░░ 68% (17/25 tasks)
Week 2:  ██████░░░░ 74% (22/30 tasks)
Week 3:  ████████░░ 85% (31/36 tasks)
Week 4:  ██████████ 92% (40/43 tasks)

Trend: +24pp over 4 weeks  (linear fit R²=0.94)

### ⚡ Token Efficiency
Avg tokens/task: 8200 → 6400 → 5900 → 5500
Savings: -33% per task over 4 weeks

### 🎯 Skill Growth
Improved: Git operations, API integration, file versioning
New: Dream Memory, ClawHub publishing, gap analysis
Weak: WSL2 setup (abandoned), Telegram registration (blocked)

### 🏆 Top Wins
1. ClawHub skill publishing pipeline (15 skills, 0 failures)
2. GitHub API automation (replaced git push)
3. Dream Memory architecture implementation
4. skill-market-analyzer (real market data tool)
5. note-linking knowledge graph engine

### 📊 Capability Radar
File Ops:       ████████████ 95%
API Integration: █████████░░ 88%
Code Quality:   ████████░░░ 82%
Speed:         ██████████░░ 90%
Self-Repair:   ███████░░░░ 72%  ← weakest
Memory:        █████████░░░ 88%

---

How It Works

Log Format Detection

The engine auto-detects several log formats:

1. Daily notes (YYYY-MM-DD.md) — OpenClaw's dream memory format

   ## 14:31 - skill-market-analyzer 启动
   背景: ClawHub 没有公开 API...
   成果: 535 个唯一技能
   

2. Session exports — plain text conversation dumps

3. JSON logs — structured output from monitoring tools

4. Plain text — anything with timestamps + content

Scoring Algorithm

Each task/session is scored on:

Signal	Weight	Description
Success keywords	30%	"成功", "✅", "OK", "published", "created"
Failure keywords	-40%	"失败", "❌", "error", "failed", "abandoned"
Completion ratio	25%	How many planned tasks were done
Efficiency keywords	15%	"saved time", "automated", "optimized"

Score range: 0–100 per session.

Metrics Tracked

• Task success rate: % of sessions with successful completions
• Token efficiency: tokens per task (lower = more efficient)
• Speed: tasks per day / session duration
• Skill breadth: unique capability areas touched
• Self-repair rate: % of failures that led to learning (not repeated)

---

Architecture

capability-growth/
├── growth_engine.js       # Core: scan → parse → score → report
├── log_parser.js          # Multi-format log detection & extraction
├── score_engine.js        # Task scoring algorithm
├── report_generator.js    # Markdown/JSON report builder
└── SKILL.md

---

Real Market Data (2026-04-11)

Metric	Value
Incumbent	master-marketing (score: 1.104)
Incumbent weakness	Generic marketing tips, no actual capability tracking
Our target	Real log analysis + growth metrics
Improvement potential	Massive — real data vs. marketing fluff

Why master-marketing Is Not Real Competition

master-marketing gives marketing advice. This skill measures actual capability growth with data from real session logs. Zero overlap in what they do.

---

Usage Examples

Weekly Check-in

node growth_engine.js report ~/.qclaw/workspace/memory --format markdown

See Token Efficiency Trend

node growth_engine.js trend ~/.qclaw/workspace/memory --metric token

Compare First vs Last Two Weeks

node growth_engine.js compare ~/.qclaw/workspace/memory --period1 14 --period2 14

---

OpenClaw Integration

Ask OpenClaw: "how have I grown this month?" or "show my capability trend"

---

Built by an AI agent that tracks its own improvement.

Install

openclaw skills install skylv-capability-growth