Skip to content
Back to skills

Runtime Logger

ASecurity

Emit info-level logs to file during task execution, similar to logger.info() calls in code

  • 666 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added September 2, 2026
ai-agentsgobashrefactoringapi

Works with

  • api

Security analysis

A100/100

Scanned September 2, 2026

npx -y skills add majiayu000/claude-skill-registry --skill runtime-logger --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Runtime Logger?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Runtime Logger
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-runtime-logger/badge)](https://www.skillsdirectory.com/skills/majiayu000-runtime-logger)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: runtime-logger
description: Emit info-level logs to file during task execution, similar to logger.info() calls in code
---

# Runtime Activity Logger

Log key activities to `.claude/logs/activity.log` using bash as you work. This creates a trace of agent reasoning and actions analogous to info-level logging in deterministic code.

## Setup

Before starting any task, ensure the log directory exists:

```bash
mkdir -p .claude/logs
```

## Log Format

Use this consistent format for all log entries:

```bash
echo "[$(date -Iseconds)] [LEVEL] phase: message" >> .claude/logs/activity.log
```

### Log Levels

| Level | Use For |
|-------|---------|
| `INFO` | Normal operations, progress updates, decisions |
| `DEBUG` | Detailed technical information when troubleshooting |
| `WARN` | Unexpected situations that don't block progress |
| `ERROR` | Failures that affect task completion |

## When to Log

### Task Lifecycle
- **Task start**: Log the goal and planned approach
- **Task completion**: Log summary, outcome, and any artifacts created

### Tool Usage
- **Before significant tool use**: Log what tool and why
- **After tool use**: Log outcome (success/failure, key findings)

### Decision Points
- **Architectural choices**: Log reasoning for design decisions
- **Branch points**: Log why one approach was chosen over alternatives
- **Assumptions**: Log any assumptions being made

### Subagent Activity
- **Delegation**: Log task being delegated and to which subagent
- **Completion**: Log subagent results when they return

## Examples

### Starting a Task
```bash
echo "[$(date -Iseconds)] [INFO] start: Beginning refactoring task for payment module" >> .claude/logs/activity.log
echo "[$(date -Iseconds)] [INFO] planning: Will analyze current structure, identify patterns, then apply Strategy pattern" >> .claude/logs/activity.log
```

### Tool Usage
```bash
echo "[$(date -Iseconds)] [INFO] tool: Reading src/payments/processor.py to understand current implementation" >> .claude/logs/activity.log
echo "[$(date -Iseconds)] [INFO] tool-result: Found 3 payment types with duplicated validation logic" >> .claude/logs/activity.log
```

### Decision Points
```bash
echo "[$(date -Iseconds)] [INFO] decision: Using Strategy pattern - cleaner than inheritance for 3 payment types" >> .claude/logs/activity.log
echo "[$(date -Iseconds)] [INFO] decision: Keeping backward compatibility by wrapping legacy interface" >> .claude/logs/activity.log
```

### Subagent Delegation
```bash
echo "[$(date -Iseconds)] [INFO] delegate: Spawning test-writer subagent for unit test coverage" >> .claude/logs/activity.log
echo "[$(date -Iseconds)] [INFO] delegate-complete: test-writer created 12 unit tests, all passing" >> .claude/logs/activity.log
```

### Warnings and Errors
```bash
echo "[$(date -Iseconds)] [WARN] Found deprecated API usage in line 142, will need migration" >> .claude/logs/activity.log
echo "[$(date -Iseconds)] [ERROR] Build failed: missing dependency 'stripe-sdk', attempting to resolve" >> .claude/logs/activity.log
```

### Task Completion
```bash
echo "[$(date -Iseconds)] [INFO] complete: Refactoring finished - 3 files modified, 12 tests added" >> .claude/logs/activity.log
echo "[$(date -Iseconds)] [INFO] artifacts: Created src/payments/strategies/*.py, updated tests/test_payments.py" >> .claude/logs/activity.log
```

## Log File Management

### Viewing Logs in Real-Time
```bash
tail -f .claude/logs/activity.log
```

### Rotating Logs
For long-running projects, consider date-based log files:
```bash
echo "[$(date -Iseconds)] [INFO] message" >> .claude/logs/activity-$(date +%Y-%m-%d).log
```

## Integration with Subagents

Subagents inherit access to bash and should follow the same logging conventions. When delegating tasks, instruct subagents to:

1. Log their start and completion
2. Use the same log file path
3. Prefix messages with their agent name for traceability

Example subagent log entry:
```bash
echo "[$(date -Iseconds)] [INFO] [test-writer] start: Generating unit tests for PaymentStrategy classes" >> .claude/logs/activity.log
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.