← Back to Skills Marketplace
Squid
by
Dominik Szopa
· GitHub ↗
· v1.0.1
· MIT-0
112
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install squid-pipeline
Description
Create, modify, and debug agentic pipelines with Squid. Define multi-agent YAML workflows with spawn (OpenClaw, Claude Code, OpenCode), gates, parallel execu...
Usage Guidance
This skill is internally coherent for authoring and testing Squid YAML pipelines, but exercise caution before executing anything: 1) Inspect the bundled example pipelines and tests (they can run shell commands, git commits, create PRs, call CLIs like docker/kubectl/gh) and confirm you understand what they will do. 2) If you follow the SKILL.md install steps, review the GitHub repo code (https://github.com/dominno/squid) before cloning/building; running npm install/build executes third-party code on your machine. 3) Use the 'sandbox' test mode and the provided .test.yaml files to validate behavior without running real commands; only run integration modes after auditing and optionally mocking dangerous steps. 4) Be mindful that executing pipelines may require external credentials (git, GitHub CLI, cloud CLIs, Slack/PagerDuty tokens); provide least-privilege credentials and avoid exposing secrets. 5) If you want extra assurance, run installs inside an isolated environment (container or VM) and review any network activity during install/run.
Capability Analysis
Type: OpenClaw Skill
Name: squid-pipeline
Version: 1.0.1
The squid-pipeline skill bundle provides a comprehensive framework for an AI agent to orchestrate complex, multi-agent workflows using the Squid pipeline engine. The instructions in SKILL.md and the reference files (step-types.md, patterns.md) emphasize safety and reliability, mandating human approval gates before side effects (Rule R1), iteration limits on loops (Rule R5), and mandatory error handling (Rule R2). The provided examples (e.g., multi-agent-dev.yaml, simple-deploy.yaml) demonstrate legitimate CI/CD and development use cases without any evidence of malicious intent, data exfiltration, or unauthorized persistence.
Capability Assessment
Purpose & Capability
The name/description (Squid pipeline authoring, multi-agent workflows) matches the included files: extensive YAML examples, references, and test schemas. The skill does not require unrelated environment variables or credentials. Examples legitimately reference CLIs (docker, kubectl, gh, openclaw, claude) because pipelines are intended to orchestrate external tools.
Instruction Scope
SKILL.md instructs the agent/user to read bundled reference docs before generating pipelines — this is appropriate and explicit. It also includes human-facing install/run commands (git clone https://github.com/dominno/squid.git, npm install, npx squid run ...). Those are installation/run instructions for the Squid tool and examples and are coherent with the skill goal, but they do instruct running network and shell actions ( cloning a GitHub repo, npm install, running CLI tools ) which will execute arbitrary code if the user follows them. The examples contain run steps that perform file/git operations and create PRs (gh pr create) — expected for a pipeline tool but they can modify repos and require external credentials if actually executed.
Install Mechanism
The registry has no formal install spec; installation instructions live in SKILL.md and point to a GitHub repository (github.com/dominno/squid) and npm build steps. Using GitHub as the source is common and reasonable, but because there is no automated, vetted install spec in the registry, following SKILL.md will clone and build third‑party code locally. That operation writes code to disk and runs npm scripts — a normal install step but higher-risk compared to an instruction-only skill that never asks you to fetch/execute external code.
Credentials
The skill declares no required env vars, binaries, or secrets. The example pipelines may require external credentials to interact with services (git, gh, cloud CLIs, Slack/PagerDuty hooks) but those are not requested or embedded by the skill itself. This is proportionate for a pipeline-orchestration skill.
Persistence & Privilege
The skill is not marked always:true and does not request to modify other skills or system-wide config. It is instruction-only and does not ask for persistent privileges. Pipelines produced by the skill may create persistent side effects when executed, but that is normal for this domain and not a registry-level privilege escalation.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install squid-pipeline - After installation, invoke the skill by name or use
/squid-pipeline - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
### Robust JSON Extraction
- New `json-extract.ts` module: `parseAgentOutput()` extracts JSON from markdown fences (`` ```json ... ``` ``), embedded objects in prose, and raw JSON
- All 3 adapters (OpenClaw, Claude Code, OpenCode) now use `parseAgentOutput()` instead of fragile `JSON.parse(stdout)`
- Fixes the production issue where OpenClaw agents return JSON wrapped in markdown or prose instead of raw JSON
- 16 unit tests covering all extraction patterns
### Claude Code Sub-Agent Support
- `agentId` now works with the `claude-code` adapter — passed as `--agent <name>` to the CLI
- Targets Claude Code sub-agents defined in `.claude/agents/<name>.md`
- `agentId` is no longer OpenClaw-only — works across `claude-code` and `openclaw` adapters
### Parser Validation
- Warning when `model:` is set with `runtime: subagent` — model is ignored for OpenClaw sub-agents (model is configured per agent in OpenClaw, not in the pipeline)
### OpenClaw Adapter Cleanup
- Removed dead `OPENCLAW_TOKEN` / `OPENCLAW_GATEWAY_TOKEN` / `CLAWD_TOKEN` env var handling — the `openclaw` CLI never reads these; it authenticates via `openclaw config` stored credentials
- Removed `token` config field from `OpenClawConfig` interface
### Documentation
- Updated `docs/adapters.md`: fixed `agentId` support table (now "Yes" for Claude Code), corrected OpenClaw auth to CLI-only (no HTTP mode), updated comparison table
- Updated SKILL.md with full adapter guide: when to use which adapter, `SQUID_AGENT` env var, `agentId` cross-adapter table, `model:` behavior per runtime
- Added raw GitHub URLs for all doc files in SKILL.md Documentation Links section
- Clarified `${ref}` interpolation in transforms (R7) vs bare `$ref` in `when:` conditions
v1.0.0
squid-pipeline 0.1 — Initial release
- Introduces Squid pipeline YAML workflow support, including multi-agent orchestration, gates, parallel execution, loops, branching, and structured approvals.
- Enforces 10 core rules for safe, deterministic, and testable pipelines (e.g., mandatory gates before side effects, onError handling, spawn output specification, retry on network, loop limits).
- Requires use of provided reference files (step-types, patterns, testing) for accurate pipeline and test generation.
- Supports agent adapters (OpenClaw, Claude Code, OpenCode), sub-pipelines, and human-in-the-loop steps.
- Provides install instructions, pipeline structure guide, and example patterns for rapid onboarding.
- Mandates comprehensive test coverage for pipelines, including happy path, rejections, failures, and restart exhaustion.
Metadata
Frequently Asked Questions
What is Squid?
Create, modify, and debug agentic pipelines with Squid. Define multi-agent YAML workflows with spawn (OpenClaw, Claude Code, OpenCode), gates, parallel execu... It is an AI Agent Skill for Claude Code / OpenClaw, with 112 downloads so far.
How do I install Squid?
Run "/install squid-pipeline" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Squid free?
Yes, Squid is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Squid support?
Squid is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Squid?
It is built and maintained by Dominik Szopa (@dominno); the current version is v1.0.1.
More Skills