← Back to Skills Marketplace
cyber-bye

Workflow Guardian

by cyber-bye · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
96
Downloads
2
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install workflow-guardian
Description
Defines, enforces, and tracks structured workflows for any task type. Manages do/don't rules, execution sequences, hard gates, soft advisories, checkpoints,...
README (SKILL.md)

Workflow Guardian Skill

Purpose

Give the agent a system for HOW to work — not just what to do. Every repeating task type gets a defined workflow. Every workflow has rules, checkpoints, and gates. Violations are caught in real time and logged for review.


Core Concepts

Concept Meaning
workflow Ordered step sequence for a specific task type
rule A do or don't that applies globally or per workflow
checkpoint Mandatory verification point inside a workflow
gate Hard stop — cannot proceed until condition is met
advisory Soft warning — log it, do not block
violation A rule or gate broken during execution
post-fix Corrective action applied after a violation

Workflow States

State Meaning
pending Defined, not yet started
active Currently executing
paused Waiting on gate condition or owner input
completed Finished successfully, all checkpoints passed
aborted Stopped — violation or owner cancel
post-fix Completed but violation found, correction in progress

Rule Types

Type Enforcement What happens on break
hard-do Mandatory action Gate — workflow stops
hard-dont Prohibited action Gate — workflow stops
soft-do Recommended action Advisory — log + warn
soft-dont Discouraged action Advisory — log + warn
global Applies to ALL workflows Either hard or soft
scoped Applies to one workflow only Either hard or soft

Folder Structure

workflow-guardian/
  workflows/
    active/                  ← currently defined workflows
      \x3Cworkflow-id>/
        workflow.md          ← definition: steps, gates, rules
        run-log.md           ← execution history (append-only)
    archived/                ← deprecated workflows (never delete)
    templates/               ← reusable workflow templates
  rules/
    global/
      do.md                  ← global hard/soft dos
      dont.md                ← global hard/soft don'ts
    do/                      ← scoped do rules by workflow
      \x3Cworkflow-id>.md
    dont/                    ← scoped don't rules by workflow
      \x3Cworkflow-id>.md
  checkpoints/
    \x3Cworkflow-id>.md         ← checkpoint definitions per workflow
  violations/
    raw/                     ← captured immediately on detection
      YYYY-MM-DD-\x3Cwf>-\x3Cslug>/entry.md
    reviewed/                ← processed violations
  hooks/
    pre-task.md              ← runs before any workflow step
    post-task.md             ← runs after any workflow step
    on-violation.md          ← fires immediately on rule break
    on-complete.md           ← fires when workflow completes
  memory/
    schema.json              ← validated memory structure
    index.json               ← runtime state (auto-managed)
  templates/
    workflow.md              ← template for new workflow definitions
    violation-entry.md       ← template for violation entries
  crons/
    active/
    completed/
  WORKFLOW_LOG.md            ← master execution log (append-only)
  RULES_INDEX.md             ← all rules in one place
  STATS.md                   ← compliance rate, violation counts
  SOUL.md                    ← persistent soul context
  AGENT.md                   ← behavioral enforcement rules

Slug Format

Workflow IDs: \x3Ccategory>-\x3Ctask-type> e.g. code-review, file-creation, api-integration Violation slugs: YYYY-MM-DD-\x3Cworkflow-id>-\x3Crule-broken>


Immediate Violation Capture Rule

The moment a rule is broken — before any other action:

  1. Write to violations/raw/\x3Cslug>/entry.md (minimum viable)
  2. Fire hooks/on-violation.md
  3. If hard rule → GATE: stop workflow, surface to owner
  4. If soft rule → ADVISORY: log, warn, continue

Post-Fix Protocol

When a violation is found AFTER workflow completion:

  1. Set workflow state to post-fix
  2. Write violation entry
  3. Determine corrective action
  4. Apply fix if autonomous, else escalate
  5. Mark workflow completed only after fix confirmed

Workflow Definition Requirements

Every workflow in workflows/active/ MUST define:

  • steps[] — ordered list with descriptions
  • gates[] — hard stop conditions
  • checkpoints[] — verification points
  • rules[] — scoped dos and don'ts
  • post-fix-policy — what to do if violation found after completion

Stats Tracking

After every workflow completion or violation:

  • Update STATS.md compliance rate
  • Update memory/index.json
  • If compliance rate drops below 80% for any workflow → advisory to owner
Usage Guidance
This skill appears coherent and local-only: it manages workflows by reading/writing files in a contained folder structure and does not request credentials or network access. Before installing, consider: 1) Back up any existing workspace files you care about (WORKFLOW_LOG.md, memory/index.json, SOUL.md) because the skill expects to append/update them. 2) Review the global rules/docs (AGENT.md, rules/global/*.md) to confirm you accept the strict gate semantics and the claim that some rules 'cannot be overridden by owner instruction' — if you need manual bypasses, plan how you'll signal overrides. 3) Test in a non-production workspace to verify behavior and that the agent's file writes match your expectations. If you need the skill to call external services or require fewer persistent writes, request a variant that limits persistence or exposes explicit owner-controlled override mechanisms.
Capability Analysis
Type: OpenClaw Skill Name: workflow-guardian Version: 1.0.0 The 'workflow-guardian' skill is a comprehensive framework for defining and enforcing structured task sequences (workflows) for an AI agent. It implements a governance system using rules, gates, and checkpoints to ensure auditability and process integrity, with state maintained across files like `SOUL.md`, `STATS.md`, and `memory/schema.json`. While it contains strong instructions for the agent to prioritize global safety rules over user overrides (e.g., requiring confirmation for destructive operations in `rules/global/dont.md`), these constraints are aligned with its stated purpose of workflow enforcement and lack any indicators of malicious intent, data exfiltration, or unauthorized system access.
Capability Assessment
Purpose & Capability
The name/description (workflow enforcement, gates, checkpoints, violation logging) matches the content of SKILL.md and the included docs. No unrelated environment variables, binaries, or external services are requested.
Instruction Scope
All runtime instructions operate on a clearly defined local folder layout (workflows/, rules/, checkpoints/, violations/, memory/, SOUL.md, logs). The skill directs the agent to read/write those files and to run hooks before/after steps. This is expected for a workflow enforcer, but note the skill prescribes global rules that 'cannot be overridden by owner instruction' and enforces gate behavior strictly — this is a behavioral policy choice that may surprise operators.
Install Mechanism
Instruction-only skill with no install spec and no code files to execute. No downloads, package installs, or binaries are required, minimizing install-time risk.
Credentials
The skill requests no environment variables, credentials, or config paths outside its own document tree. All state is local (files and memory/schema.json). No secret exfiltration vectors are requested in the manifest or instructions.
Persistence & Privilege
The skill writes persistent state (SOUL.md, memory/index.json, WORKFLOW_LOG.md, run logs, violations). It does not set always:true and does not request system-wide config changes, but it does expect to maintain persistent 'soul' context and runtime state — ensure you are comfortable with the agent writing and updating these files in the workspace.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install workflow-guardian
  3. After installation, invoke the skill by name or use /workflow-guardian
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Initial release: full workflow definition, do/dont rule system, gates, checkpoints, violations, post-fix protocol, soul + memory integration
Metadata
Slug workflow-guardian
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Workflow Guardian?

Defines, enforces, and tracks structured workflows for any task type. Manages do/don't rules, execution sequences, hard gates, soft advisories, checkpoints,... It is an AI Agent Skill for Claude Code / OpenClaw, with 96 downloads so far.

How do I install Workflow Guardian?

Run "/install workflow-guardian" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Workflow Guardian free?

Yes, Workflow Guardian is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Workflow Guardian support?

Workflow Guardian is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Workflow Guardian?

It is built and maintained by cyber-bye (@cyber-bye); the current version is v1.0.0.

💬 Comments