How to Install and Use Sag in OpenClaw

Open your terminal and run the clawhub install command to add Sag to your OpenClaw instance. ClawHub downloads the sag skill package, verifies its checksum against the registry, and registers it with your local OpenClaw configuration. Installation takes 10-20 seconds. When it completes, the terminal shows a confirmation message with the installed version and any configuration keys that may be required. Sag's core pattern library may work without additional configuration, though some patterns that interact with external data sources may require credentials. After installation, run `clawhub info sag` to see the full documentation including the list of built-in automation patterns, their trigger phrases, and any pattern-specific configuration options. Understanding the pattern library before using Sag helps you write more effective prompts.

1# Install sag
2clawhub install sag
3
4# Verify installation
5clawhub list
6
7# View pattern library and configuration reference
8clawhub info sag

Before running automation patterns, review the output of `clawhub info sag` to understand what patterns are available and how they are invoked. Each pattern in Sag has a name, a description, and specific input/output expectations — understanding these before using the skill produces much more accurate results than trial and error. If Sag requires any configuration (such as default output formats, pattern-specific settings, or credentials for external data source patterns), add these to your OpenClaw config using `clawhub config set` or by editing `~/.openclaw/skills/sag.yaml`. Apply the changes with `clawhub reload`. For Sag patterns that process external data, the relevant data must be provided directly in your OpenClaw prompt or sourced from another skill that Sag can chain with. Sag is a pattern execution engine — it applies defined logic to inputs you provide, so the quality of your inputs directly affects the quality of outputs.

1# Review pattern library
2clawhub info sag
3
4# Configure Sag in ~/.openclaw/skills/sag.yaml
5sag:
6  default_output_format: "structured"  # structured | plain | json
7  pattern_verbosity: "normal"          # silent | normal | verbose
8  max_batch_size: 100                  # max items in batch operations
9  strict_pattern_matching: false       # require exact pattern names (default: fuzzy)
10
11# Apply configuration
12clawhub reload
13
14# Verify Sag is active
15clawhub status sag

Open OpenClaw chat and test Sag by invoking one of its built-in automation patterns. Start with a simple, self-contained pattern that does not require external data sources — something where you can provide all the input directly in your prompt and immediately verify the output is correct. Good first tests are data cleaning or validation patterns: give Sag a list of items with deliberate errors and ask it to apply a specific cleaning pattern. The result should be a consistently structured output that matches the pattern's defined behavior. Invoke Sag explicitly by name to ensure it handles the prompt: 'Sag, apply [pattern name] to...' or 'Using Sag, run the [pattern name] pattern on...'. Once you are familiar with how OpenClaw routes to Sag naturally, the explicit invocation becomes optional for prompts that clearly match Sag's pattern vocabulary. RapidDev can help teams define custom Sag automation patterns for specific operational workflows — particularly useful when the built-in pattern library does not cover your exact use case.

While Sag ships with a built-in pattern library, you can also define custom patterns in the sag.yaml configuration file for workflows specific to your use case. Custom patterns follow the same structure as built-in patterns — a name, a description, input expectations, a sequence of processing steps, and an output format. Custom patterns are particularly valuable for recurring operational workflows that are specific to your team or project. Instead of describing the full logic in every prompt, you define it once in the configuration file and then invoke it by name from OpenClaw chat. After defining a custom pattern, run `clawhub reload` to apply the configuration and then test the pattern with a representative input to verify it behaves as expected. Complex patterns may require iteration — refining the step definitions after seeing how Sag interprets and executes them.

1# Define custom patterns in ~/.openclaw/skills/sag.yaml
2sag:
3  custom_patterns:
4    vendor-contact-cleanup:
5      description: "Standardize and validate vendor contact records"
6      input_format: "list of contact records"
7      steps:
8        - normalize_phones: "format to +1-XXX-XXX-XXXX"
9        - capitalize_companies: "title case for company names"
10        - validate_email_domain: "flag where domain doesn't match company name"
11      output_format: "structured_report"
12      report_sections:
13        - passed: "records that passed all rules"
14        - flagged: "records with validation issues"
15    data-quality-check:
16      description: "Run standard data quality checks on a dataset"
17      steps:
18        - check_completeness: "flag empty required fields"
19        - check_format_consistency: "flag format outliers"
20        - check_duplicates: "identify duplicate entries"
21      output_format: "summary_with_counts"

Sag's automation patterns become most powerful when chained with other ClawHub skills in a multi-step OpenClaw workflow. You might use ByteRover to fetch raw data, Sag to apply a validation or transformation pattern to that data, and Supermemory to save the processed result for later retrieval — all in a single OpenClaw conversation. Cross-skill pipelines in OpenClaw work by describing the full workflow intent in a single multi-step prompt. OpenClaw routes each component to the appropriate skill and passes outputs between steps. Describing the pipeline clearly — including what data flows from one step to the next — is the key to getting reliable multi-skill execution. For regularly run pipelines, consider whether the workflow could be encapsulated as a named Larry workflow template that invokes Sag patterns at specific steps. Larry handles the orchestration and interaction layer, while Sag handles the structured processing steps within the pipeline.