file-ops

name	file-ops
version	1.0.0
description	File system operations for the builder agent: create directories, write files, copy templates, read file contents. Used when generating multi-agent system output directories.
user-invocable	false
metadata	{"openclaw":{"requires":{"bins":["mkdir","cp","cat","sed"]}}}

File Operations Skill

Provides file system primitives for the builder agent to create generated system directories.

Generation Modes

openclaw-native (default)

Generate OpenClaw-native agents:

• Skills: Inline exec commands in SKILL.md
• Orchestration: Lobster workflow + cron/*.json
• No run.py files, no main.py

standalone-pipeline (legacy)

Generate standalone automation pipelines:

• Skills: SKILL.md + run.py subprocess scripts
• Orchestration: main.py orchestrator
• Deployable Python pipelines

Current mode: openclaw-native

Operations

Create Directory Structure (OpenClaw-Native)

mkdir -p output/<system-name>/{cron,workflows,workspace,skills}
mkdir -p output/<system-name>/skills/{data-ingestion-openclaw,<custom-skill-1>,<custom-skill-2>}

Write File

Write content to a file path. Use heredoc for multi-line content:

cat > output/<system-name>/openclaw.json << JSONEOF
{content}
JSONEOF

Copy Template (OpenClaw-Native)

Copy template and substitute variables:

sed -e "s|{{SKILL_ID}}|<skill-id>|g" \
    -e "s|{{SKILL_NAME}}|<skill-name>|g" \
    -e "s|{{EXEC_COMMANDS}}|<commands>|g" \
    templates/openclaw-native/skill-inline-exec.md.template > \
    output/<system-name>/skills/<skill-id>/SKILL.md

Template Variables (OpenClaw-Native)

skill-inline-exec.md.template:

• {{SKILL_ID}}: Skill identifier (kebab-case)
• {{SKILL_NAME}}: Human-readable name
• {{SKILL_DESCRIPTION}}: Short description
• {{REQUIRED_BINS}}: JSON array of required binaries
• {{REQUIRED_ENV_VARS}}: JSON array of env var names
• {{PRIMARY_ENV_VAR}}: Main env var (e.g., API key)
• {{EXEC_COMMANDS}}: Bash/python commands to execute
• {{EXPECTED_OUTPUT}}: What the skill produces
• {{ERROR_HANDLING}}: Error handling logic

cron-job.json.template:

• {{JOB_NAME}}: Human-readable job name
• {{CRON_EXPRESSION}}: Cron syntax (e.g., "30 2 * * *")
• {{TIMEZONE}}: Timezone (default: "UTC")
• {{TRIGGER_MESSAGE}}: Message sent to agent on trigger
• {{MODEL}}: LLM model (default: "openrouter/anthropic/claude-sonnet-4.5")
• {{TIMEOUT_SECONDS}}: Job timeout (default: 600)

soul-workflow.md.template:

• {{AGENT_NAME}}: Agent name
• {{AGENT_TYPE}}: Type (e.g., "task prioritization assistant")
• {{AGENT_PURPOSE}}: Purpose statement
• {{TONE}}: Tone description
• {{SKILL_LIST}}: Bullet list of available skills
• {{TRIGGER_MESSAGE}}: Cron trigger message
• {{WORKFLOW_STEPS}}: Numbered list of execution steps
• {{DELIVERY_METHOD}}: How results are delivered
• {{ALERT_CHANNEL}}: Error alert channel
• {{MANUAL_COMMANDS}}: User commands supported

Read File

cat output/<system-name>/openclaw.json

List Directory

ls -la output/<system-name>/

Output Directory Convention

All generated systems are created under output/<system-name>/ relative to the factory workspace root. The builder MUST use this convention.

OpenClaw-Native File Structure

output/<system-name>/
├── openclaw.json
├── README.md
├── .env.example
├── .gitignore
├── check-environment.sh         ← Generated from template
├── install-dependencies.sh      ← Generated from template
├── test-workflow.sh             ← Generated from template
├── cron/
│   └── <job-name>.json
├── workflows/
│   └── main.yaml
├── workspace/
│   ├── SOUL.md
│   └── IDENTITY.md
└── skills/
    ├── data-ingestion-openclaw/SKILL.md
    └── <custom-skill>/SKILL.md

Helper Script Generation

The builder MUST generate three executable helper scripts for every system:

1. check-environment.sh: Validates binaries, env vars, API connectivity
2. install-dependencies.sh: Auto-installs system packages and npm packages
3. test-workflow.sh: Tests the data pipeline manually

See templates/openclaw-native/README-SCRIPTS.md for template usage.