AgentSkillsCN

plan-tf-resource

为 CrowdStrike 提供程序中的全新 Terraform 资源或数据源规划并创建 GitHub 问题

SKILL.md
--- frontmatter
name: plan-tf-resource
description: Plan and create a GitHub issue for a new Terraform resource or data source in the CrowdStrike provider
disable-model-invocation: true
allowed-tools: Bash, Read, AskUserQuestion

Plan Terraform Resource/Data Source

Plan and create a GitHub issue for a new Terraform resource or data source in the CrowdStrike provider.

Process Overview

  1. •Initial questions → API discovery → Schema design → Issue generation → User review → Create issue

Phase 1: Initial Questions

Step 1: Initial Information Gathering

Present this question as text output (do not use AskUserQuestion tool):

"What CrowdStrike feature or capability do you want to add to the provider? If you know the gofalcon package name, please include it."

Wait for the user to provide:

  • •Feature name or description (e.g., "response_policy" or "manages real-time response policies")
  • •Optionally: gofalcon package name (e.g., "response_policies")

Step 2: Infer or Ask Resource Type

Based on the user's description:

  • •If it's clear this is a resource (manages/creates/updates something) or data source (reads existing data), proceed with that understanding
  • •If unclear, ask using AskUserQuestion:
    • •Is this a resource or data source?
    • •Options: "Resource" or "Data Source"

Phase 1.5: Package Validation/Discovery

Always perform package validation or discovery, regardless of whether the user provided a package name.

Scenario A: User Provided Package Name

If the user provided a gofalcon package name in Phase 1, validate it:

  1. •Verify package exists:
bash
go doc github.com/crowdstrike/gofalcon/falcon/client/<package>
  1. •Check for required CRUD operations: Look for methods matching these patterns (at least Create and Get should exist):
  • •Create operations: *Create*, *Post*
  • •Read operations: *Get*, *Query*
  • •Update operations: *Update*, *Patch*
  • •Delete operations: *Delete*
  1. •If validation fails (package doesn't exist or lacks critical operations):
  • •Inform the user: "The package <package> doesn't exist or lacks required CRUD operations."
  • •Proceed to Scenario B (discovery)
  1. •If validation passes:
  • •Proceed to Phase 2 (API Discovery)

Scenario B: User Did NOT Provide Package Name OR Validation Failed

Help discover the correct package:

  1. •Search for relevant packages using keywords from the feature description:
bash
go list github.com/crowdstrike/gofalcon/... | grep <keyword>
  1. •

    Show results to the user with brief package descriptions (if available)

  2. •

    Ask user to choose the correct package or provide additional search terms

  3. •

    Once package is identified, validate it using Scenario A steps

  4. •

    After successful validation, proceed to Phase 2 (API Discovery)

Phase 2: API Discovery

Use the following commands to learn more about the api:

bash
# Find relevant packages
go list github.com/crowdstrike/gofalcon/... | grep <keyword>

# Explore package operations
go doc <package>

# View client methods
go doc <package>.Client

# View API models
go doc github.com/crowdstrike/gofalcon/falcon/models.<Model>

Document:

  • •gofalcon package: Full package path
  • •Available operations: List CRUD methods with brief descriptions
    • •Example: EntitiesContentPatternCreate - Create content pattern
  • •Models: Main request/response model names

Phase 3: Intelligent Schema Design

Use pattern-based heuristics to automatically design the complete schema, then present it to the user for feedback.

Note: In this phase, present the schema design and questions as regular text output and wait for user responses. Do not use the AskUserQuestion tool.

Step 1: Examine API Model Fields

Use go doc to view the API model structure and field documentation:

bash
go doc github.com/crowdstrike/gofalcon/falcon/models.<ModelName>

Extract:

  • •Field names and types
  • •Field descriptions from gofalcon documentation (if available)
  • •Any enum values or constraints mentioned

Step 2: Apply Automatic Field Classification

For each field in the API model, apply these pattern-matching rules:

By Field Name Patterns:

  • •id → Computed string with UseStateForUnknown() plan modifier
  • •last_updated → Computed string (timestamp)
  • •enabled → Optional bool with booldefault.StaticBool(false) default
  • •description → Optional string with StringNotWhitespace() validator
  • •name → Required string with StringNotWhitespace() validator

By Field Type from API:

  • •Boolean fields → Usually Optional with booldefault.StaticBool(false) default
  • •String ID fields → Usually Optional unless they're discriminators
  • •Arrays in API → Optional Set or List type in Terraform
  • •Objects in API → Optional nested Object type in Terraform
  • •Enums → Add OneOf() validator with valid values

Automatic Validator Assignment:

  • •Optional strings → stringvalidator.NotEmptyOrWhitespace() or validators.StringNotWhitespace()
  • •Enum fields → stringvalidator.OneOf(...values from API docs...)
  • •Email fields → validators.StringIsEmailAddress()
  • •Collections → setvalidator.ValueStringsAre(validators.StringNotWhitespace()) for string elements
  • •Numeric ranges → int64validator.Between(min, max) if constraints mentioned in docs

Automatic Description Generation:

  1. •Primary source: Use field description from gofalcon go doc output if available as a starting point. Reword to be a good description.
  2. •Fallback patterns if gofalcon docs don't provide description:
    • •Simple fields: "The <field_name> of the <resource>."
    • •Enum fields: "The <field_name> of the <resource>. One of: <values>."
  3. •Always use MarkdownDescription format with backticks for code values and enums

Step 3: Present Complete Schema Design

After applying pattern matching to all fields, present the complete schema design:

Pattern-Based Schema Design:

I've analyzed the API model and designed the following schema using common patterns:

Assumptions made:

  • •[List key assumptions, e.g., "enabled defaults to false", "name is required with non-empty validator"]
  • •[Note any fields where pattern confidence is low]

Required attributes:

  • •field_name (type) - Description [Pattern: reason]
  • •...

Optional attributes:

  • •field_name (type) - Description [Pattern: reason]
  • •[List default values where applicable]
  • •...

Hardcoded values (not exposed to users):

  • •field_name = "value" (set internally) [Pattern: reason]
  • •...

Computed attributes (read-only):

  • •field_name (type) - Description [Pattern: reason]
  • •...

Fields needing verification:

  • •field_name - [Explain why this field is ambiguous or needs confirmation]
  • •...

Validation requirements (if applicable):

  • •Document any feature prerequisites (e.g., "Feature X must be enabled for Feature Y to work")
  • •Include these dependencies in:
    • •Attribute descriptions (e.g., "Required for all X commands")
    • •ValidateConfig requirements in the issue
  • •Use proper display names in validation requirements

Present the schema and say:

"I've designed the schema based on common patterns. Does this look correct? What would you like to change?"

STOP and wait for user response.

Iteration Process:

  1. •

    User provides feedback - They can:

    • •Approve the schema as-is (e.g., "looks good", "proceed", "that's correct") → Proceed to Phase 4
    • •Request changes to specific fields → Continue to step 2
    • •Ask questions about the patterns used → Clarify and continue to step 2
  2. •

    Apply requested changes to the schema

  3. •

    Present the updated schema (showing what changed)

  4. •

    Ask again: "Does this look correct now? Any other changes?"

  5. •

    Repeat steps 1-4 until the user explicitly approves the schema

  6. •

    Only after user approval, proceed to Phase 4 (Additional Information)

Phase 4: Additional Information

Ask the user:

  1. •

    "What are the required API scopes?" (ONLY if not already established during API discovery)

    • •If API scopes were already identified, skip this question
    • •Example: "Data Protection | Read & Write"
    • •Format: <Scope Name> | Read & Write or <Scope Name> | Read etc.
  2. •

    "Are there any questions to investigate during implementation?" (Only if there are genuinely ambiguous scenarios)

    • •Only for things that cannot be known until API testing
    • •Example: "Platform support - UI shows Windows/Mac but API doesn't accept platform"
    • •NOT for schema design decisions already made
    • •Skip this question if there are no unknowns

Phase 5: Issue Generation

CRITICAL: What NOT to Include in GitHub Issues

GitHub issues are planning documents, NOT implementation guides. DO NOT include:

  • •❌ Code for .Wrap() method - No conversion logic
  • •❌ Code for CRUD methods (.Create(), .Read(), .Update(), .Delete())
  • •❌ References to other resources as implementation examples
  • •❌ Helper function code or patterns
  • •❌ API client call examples or error handling code
  • •❌ Sections like "API Helper Functions" or "Error Handling Implementation"

DO include:

  • •✅ Complete schema definition (with validators and plan modifiers) - This is critical
  • •✅ Realistic HCL examples showing how users will use the resource
  • •✅ WHAT needs to be done (checklist of tasks)
  • •✅ API caveats discovered during planning (character limits, field transformations, special behaviors)
  • •✅ High-level notes about special handling requirements

The schema definition IS implementation code and should be complete. Everything else should be guidance, not code.

Reference Format

Before generating the issue, review examples/ioa-exclusion-issue.md to understand the expected format and level of detail.

Key characteristics:

  • •Realistic HCL example (single scenario preferred, multiple only if fundamentally different usage patterns)
  • •Complete schema Go code (validators, plan modifiers) without imports or package declaration
  • •High-level checklist (what to do, not how)
  • •Notes section for API quirks and discoveries
  • •NO .Wrap() code, NO helper functions, NO method implementations

Issue Structure

Generate a complete GitHub issue with the following structure:

Title Format

  • •Resources: Add <resource_name> resource
  • •Data Sources: Add <data_source_name> data source

Body Structure

markdown
<Brief description (1-2 sentences) of what this resource/data source manages>

## Example

```hcl
<Generate a realistic, complete Terraform HCL example showing how users will use this resource>
<Include inline comments only where they add clarity about non-obvious usage>
<Only include multiple scenarios if there are fundamentally different ways to use the resource (e.g., specific vs global configurations)>

API Details

gofalcon package: github.com/crowdstrike/gofalcon/falcon/client/<package>

Available operations:

  • •OperationName - Brief description
  • •OperationName - Brief description
  • •...

Models:

  • •Request: models.<RequestModel>
  • •Response: models.<ResponseModel>

Resource Schema

Required attributes:

  • •field_name (type) - Description
  • •...

Optional attributes:

  • •field_name (type) - Description
  • •...

Computed attributes (read-only):

  • •field_name (type) - Description
  • •...

Schema Implementation

go
func (r *<resource>Resource) Schema(ctx context.Context, req resource.SchemaRequest, resp *resource.SchemaResponse) {
    resp.Schema = schema.Schema{
        MarkdownDescription: utils.MarkdownDescription(
            "<Category>",
            "<Description>",
            apiScopesReadWrite,
        ),
        Attributes: map[string]schema.Attribute{
            "id": schema.StringAttribute{
                Computed:    true,
                Description: "The ID of the <resource>.",
                PlanModifiers: []planmodifier.String{
                    stringplanmodifier.UseStateForUnknown(),
                },
            },
            <Generate all other attributes with full validators and plan modifiers>
        },
    }
}

Implementation Checklist

Note: Checklist items and sub-bullets are high-level by default. Add indented sub-bullets under ANY item to document special handling requirements discovered during planning (e.g., "Pass nil for field X", "Validate Y cannot be combined with Z"). Do NOT include implementation code.

Resource implementation:

  • • Create package directory internal/<package>/
  • • Implement Schema() method
  • • Implement Create() method
  • • Implement Read() method
  • • Implement Update() method
  • • Implement Delete() method
  • • Implement ImportState() method
  • • Add ValidateConfig() method
  • • Add .Wrap() method to convert API response to Terraform model
  • • Register resource in internal/provider/provider.go

Testing and documentation:

  • • Create acceptance tests in internal/<package>/<file>_test.go
  • • Create example in examples/resources/<resource_name>/resource.tf
  • • Create import script in examples/resources/<resource_name>/import.sh
  • • Run make gen to generate documentation

Required API Scopes

  • •<Scope Name> | <Permissions>

Testing Notes

  • •Use resource.ParallelTest() for concurrent execution
  • •Test full lifecycle: create, read, update, import, destroy <Add specific test scenarios if applicable>

Notes

<Only include this section if there are API quirks, caveats, or special cases discovered during planning>

<Quirk Category>: <Description of the quirk and how it should be handled>

<Add additional notes as needed> ```

Phase 6: User Review and Creation

CRITICAL: Show the complete issue title and body to the user.

Say: "Here's the GitHub issue I've prepared. Please review and let me know if you'd like any changes before I create it."

Wait for user feedback. They can:

  • •Approve: Proceed to create the issue
  • •Request changes: Update the issue and show again
  • •Provide specific edits: Apply changes and show updated version

After user approval, create the issue using proper heredoc syntax to avoid escaped backticks:

bash
gh issue create --label "enhancement" --title "title here" --body "$(cat <<'EOF'
body here
EOF
)"

Return the issue URL to the user.

Reference

Common Validators

  • •int64validator.AtLeast(N) - Minimum value constraint
  • •stringvalidator.NotEmptyOrWhitespace() - No empty/whitespace strings
  • •stringvalidator.OneOf("val1", "val2") - Enum value validation
  • •Custom validators - Use pseudo-code with TODO comments

Common Plan Modifiers

  • •stringplanmodifier.UseStateForUnknown() - For computed ID fields
  • •stringplanmodifier.RequiresReplace() - Force resource recreation on change

Example Custom Validator Pseudo-code

go
Validators: []validator.String{
    // TODO: Implement custom regex compilation validator
    validators.ValidRegexPattern(),
},

Important Notes

Issue Philosophy

  • •Issues define WHAT needs to be built, not HOW to build it
  • •The implementer already knows provider patterns - don't tell them how to code
  • •Focus on the unique aspects of THIS resource (schema, API quirks, special cases)
  • •The schema is the contract - make it complete and correct

What Makes a Good Issue

  • •Schema first: The schema definition is the most important part - it sets up everything
  • •Document discoveries: Note any API quirks found during planning (limits, transformations, unexpected behaviors)
  • •High-level checklist: List what needs to be done, let implementer decide how
  • •No code examples: Except the schema - everything else is prescriptive

Common Mistakes to Avoid

  • •Showing how to implement .Wrap() - implementer knows this pattern
  • •Referencing other resources - don't create dependencies on understanding other code
  • •Including helper function pseudo-code - not needed in planning document
  • •Being too specific about file organization - let implementer structure the package

General Guidelines

  • •Never reference existing resources as implementation examples
  • •Always explain the assumptions made when presenting the schema design
  • •Always show issue content before creating
  • •Use gofalcon documentation as the primary source for field descriptions