fabro/test/attractor/green_test_moderate.dot

digraph linkcheck {
    graph [
        goal="Build a Go CLI tool that crawls URLs, checks links for HTTP status, and reports broken links with robots.txt support",
        rankdir=LR,
        default_max_retries=3,
        retry_target="impl_setup",
        model_stylesheet="
            * { model: gpt-5.2-codex;reasoning_effort: medium; }
            .hard { model: gpt-5.2-codex;reasoning_effort: high; }
            .verify { model: gpt-5.2-codex;reasoning_effort: medium; }
            .review { model: gpt-5.2-codex;reasoning_effort: high; }
        "
    ]

    start [shape=Mdiamond, label="Start"]
    exit  [shape=Msquare, label="Exit"]

    // Phase 1: Requirements expansion
    expand_spec [
        shape=box,
        timeout=600,
        prompt="Given the requirements: Build a Go CLI tool called 'linkcheck' that takes a URL as input, crawls the page, finds all links, checks each link for HTTP status, and outputs a report of broken links (4xx/5xx). It should respect robots.txt, have a configurable crawl depth (default 1), and output in both human-readable and JSON formats.

Expand this into a detailed spec covering:
- CLI interface (flags, arguments, usage)
- Package structure (cmd/, pkg/)
- Core types and interfaces (URL, Link, Report, etc.)
- Robots.txt parsing and compliance
- HTTP client configuration (timeouts, retries)
- Crawl algorithm (depth limits, visited tracking)
- Link extraction from HTML
- Status code categorization (2xx=ok, 4xx/5xx=broken)
- Output formats (human-readable table, JSON)
- Error handling strategy
- Test plan for each component

Write the expanded spec to .ai/spec.md.

Write status.json: outcome=succeeded"
    ]

    // Phase 2: Project setup
    impl_setup [
        shape=box,
        timeout=600,
        max_retries=2,
        prompt="Read .ai/spec.md.

Create the Go project structure:
- go.mod (module name: linkcheck)
- cmd/linkcheck/main.go (stub with basic CLI parsing)
- pkg/crawler/ directory
- pkg/robotstxt/ directory
- pkg/checker/ directory
- pkg/report/ directory

Run: go build ./...

Write status.json: outcome=succeeded if the project builds, outcome=failed with failure_reason otherwise."
    ]

    verify_setup [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify project setup.

Run:
1. go build ./...
2. go vet ./...

Check that:
- go.mod exists with correct module name
- cmd/linkcheck/main.go exists
- pkg/ directories are created

Write results to .ai/verify_setup.md.
Write status.json: outcome=succeeded if all checks pass, outcome=failed with details otherwise."
    ]

    check_setup [shape=diamond, label="Setup OK?"]

    // Phase 3: Core types and interfaces
    impl_types [
        shape=box,
        timeout=900,
        max_retries=2,
        prompt="Read .ai/spec.md.

Implement core types and interfaces per the spec.

Create pkg/crawler/types.go with:
- URL type
- Link type (URL, source, depth, status code)
- CrawlConfig struct (max depth, user agent, timeout)
- Crawler interface

Create pkg/report/types.go with:
- Report struct (total links, broken links, results slice)
- LinkResult struct (URL, status, error)
- Formatter interface

Include comprehensive documentation for all exported types.

Run: go build ./...

Write status.json: outcome=succeeded if builds, outcome=failed with failure_reason otherwise."
    ]

    verify_types [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify core types implementation.

Run:
1. go build ./...
2. go vet ./...

Check that:
- pkg/crawler/types.go defines all required types
- pkg/report/types.go defines report types
- All types have godoc comments
- No compilation errors

Write results to .ai/verify_types.md.
Write status.json: outcome=succeeded if all pass, outcome=failed with details otherwise."
    ]

    check_types [shape=diamond, label="Types OK?"]

    // Phase 4: Robots.txt parser
    impl_robotstxt [
        shape=box,
        class="hard",
        timeout=1200,
        max_retries=2,
        prompt="Read .ai/spec.md section on robots.txt compliance.

Implement robots.txt parser in pkg/robotstxt/:
- parser.go: Parse robots.txt files per RFC 9309
- matcher.go: Check if a URL is allowed for a given user-agent
- cache.go: Cache parsed robots.txt files to avoid repeated fetches

Handle:
- User-agent matching (specific agent, wildcards, default *)
- Allow/Disallow rules with path matching
- Crawl-delay directive
- Missing robots.txt (allow all)
- Malformed robots.txt (be permissive)

Create pkg/robotstxt/parser_test.go with tests for:
- Various robots.txt formats
- User-agent matching
- Path matching edge cases

Read: pkg/crawler/types.go for interfaces.

Run: go test ./pkg/robotstxt/...

Write status.json: outcome=succeeded if all tests pass, outcome=failed with failure_reason otherwise."
    ]

    verify_robotstxt [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify robots.txt parser implementation.

Run:
1. go build ./...
2. go vet ./pkg/robotstxt/...
3. go test ./pkg/robotstxt/... -v

Check test coverage:
4. go test ./pkg/robotstxt/... -cover

Write results to .ai/verify_robotstxt.md.
Write status.json: outcome=succeeded if all pass and coverage > 70%, outcome=failed with details otherwise."
    ]

    check_robotstxt [shape=diamond, label="Robots.txt OK?"]

    // Phase 5: HTTP checker
    impl_checker [
        shape=box,
        timeout=1200,
        max_retries=2,
        prompt="Read .ai/spec.md section on HTTP checking.

Implement HTTP link checker in pkg/checker/:
- checker.go: Check URL status codes
- Configure HTTP client with reasonable timeouts
- Follow redirects (up to a limit)
- Categorize status codes (2xx/3xx=ok, 4xx/5xx=broken)
- Handle network errors gracefully
- Support HEAD requests (fallback to GET if HEAD fails)

Create pkg/checker/checker_test.go with:
- Tests for various HTTP status codes
- Mock HTTP server for testing
- Timeout handling tests
- Redirect handling tests

Read: pkg/crawler/types.go for types.

Run: go test ./pkg/checker/...

Write status.json: outcome=succeeded if all tests pass, outcome=failed with failure_reason otherwise."
    ]

    verify_checker [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify HTTP checker implementation.

Run:
1. go build ./...
2. go vet ./pkg/checker/...
3. go test ./pkg/checker/... -v

Check:
- Tests cover success, failure, and error cases
- No race conditions: go test ./pkg/checker/... -race

Write results to .ai/verify_checker.md.
Write status.json: outcome=succeeded if all pass, outcome=failed with details otherwise."
    ]

    check_checker [shape=diamond, label="Checker OK?"]

    // Phase 6: HTML link extractor
    impl_extractor [
        shape=box,
        timeout=1200,
        max_retries=2,
        prompt="Read .ai/spec.md section on link extraction.

Implement HTML link extractor in pkg/crawler/:
- extractor.go: Parse HTML and extract all links
- Find links in <a href>, <link href>, <img src>, <script src>
- Convert relative URLs to absolute using base URL
- Filter out non-http(s) schemes (mailto:, javascript:, etc.)
- Handle malformed HTML gracefully
- Use golang.org/x/net/html for parsing

Create pkg/crawler/extractor_test.go with:
- Tests for relative URL resolution
- Tests for various HTML structures
- Tests for malformed HTML
- Tests for filtering non-http links

Read: pkg/crawler/types.go for types.

Run: go test ./pkg/crawler/...

Write status.json: outcome=succeeded if all tests pass, outcome=failed with failure_reason otherwise."
    ]

    verify_extractor [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify HTML extractor implementation.

Run:
1. go build ./...
2. go vet ./pkg/crawler/...
3. go test ./pkg/crawler/... -v

Check:
- All test cases pass
- Relative URL conversion is correct
- Non-http schemes are filtered

Write results to .ai/verify_extractor.md.
Write status.json: outcome=succeeded if all pass, outcome=failed with details otherwise."
    ]

    check_extractor [shape=diamond, label="Extractor OK?"]

    // Phase 7: Crawler algorithm
    impl_crawler [
        shape=box,
        class="hard",
        timeout=1500,
        max_retries=2,
        prompt="Read .ai/spec.md section on crawling algorithm.

Implement the main crawler in pkg/crawler/:
- crawler.go: Orchestrate crawling with depth limits
- Track visited URLs to avoid cycles
- Respect robots.txt using pkg/robotstxt
- Extract links using extractor.go
- Check each link using pkg/checker
- Implement breadth-first crawl up to configured depth
- Handle concurrent checking with rate limiting
- Collect results into a report

Key algorithm:
1. Start with seed URL at depth 0
2. Fetch and parse page
3. Extract all links
4. For each link:
   - Check if allowed by robots.txt
   - Check if already visited
   - Check HTTP status
   - If depth < max_depth and status ok, add to queue
5. Continue until queue empty or max depth reached

Create pkg/crawler/crawler_test.go with:
- Tests for depth limiting
- Tests for visited tracking
- Tests for robots.txt integration
- Mock HTTP responses for testing

Read: pkg/crawler/types.go, pkg/crawler/extractor.go, pkg/robotstxt/, pkg/checker/ for dependencies.

Run: go test ./pkg/crawler/...

Write status.json: outcome=succeeded if all tests pass, outcome=failed with failure_reason otherwise."
    ]

    verify_crawler [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify crawler implementation.

Run:
1. go build ./...
2. go vet ./pkg/crawler/...
3. go test ./pkg/crawler/... -v
4. go test ./pkg/crawler/... -race

Check:
- Depth limiting works correctly
- No infinite loops
- Robots.txt is respected
- No race conditions

Write results to .ai/verify_crawler.md.
Write status.json: outcome=succeeded if all pass, outcome=failed with details otherwise."
    ]

    check_crawler [shape=diamond, label="Crawler OK?"]

    // Phase 8: Output formatters
    impl_report [
        shape=box,
        timeout=900,
        max_retries=2,
        prompt="Read .ai/spec.md section on output formats.

Implement output formatters in pkg/report/:
- formatter.go: Format interface
- text.go: Human-readable table format showing broken links
- json.go: JSON format with all details

Human-readable format should show:
- Summary: X total links, Y broken
- Table of broken links: URL, Status Code, Source Page
- Color coding (red for errors) if terminal supports it

JSON format should output:
- Structured data with all checked links
- Status codes, timestamps, error messages
- Easy to parse for automation

Create pkg/report/formatter_test.go with:
- Tests for both formats
- Tests for empty results
- Tests for large result sets

Read: pkg/report/types.go for types.

Run: go test ./pkg/report/...

Write status.json: outcome=succeeded if all tests pass, outcome=failed with failure_reason otherwise."
    ]

    verify_report [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify report formatting implementation.

Run:
1. go build ./...
2. go vet ./pkg/report/...
3. go test ./pkg/report/... -v

Check:
- Both text and JSON formats work
- Output is properly formatted
- All test cases pass

Write results to .ai/verify_report.md.
Write status.json: outcome=succeeded if all pass, outcome=failed with details otherwise."
    ]

    check_report [shape=diamond, label="Report OK?"]

    // Phase 9: CLI integration
    impl_cli [
        shape=box,
        timeout=1200,
        max_retries=2,
        prompt="Read .ai/spec.md section on CLI interface.

Implement the CLI in cmd/linkcheck/main.go:
- Use flag package for argument parsing
- Required argument: URL to check
- Optional flags:
  --depth int (default 1): maximum crawl depth
  --format string (default text): output format (text or json)
  --user-agent string (default linkcheck/1.0): HTTP user agent
  --timeout int (default 10): HTTP timeout in seconds
  --verbose: enable verbose logging
- Wire together: crawler, checker, robots.txt, formatter
- Print results to stdout
- Exit code 0 if no broken links, 1 if broken links found, 2 on error

Usage example:
  linkcheck https://example.com
  linkcheck --depth 2 --format json https://example.com

Read: all pkg/ directories for integration.

Run: go build ./cmd/linkcheck

Write status.json: outcome=succeeded if builds, outcome=failed with failure_reason otherwise."
    ]

    verify_cli [
        shape=box,
        class="verify",
        timeout=300,
        prompt="Verify CLI integration.

Run:
1. go build ./cmd/linkcheck
2. ./cmd/linkcheck --help (check help text)
3. Test with a real URL: ./cmd/linkcheck https://example.com
4. Test JSON output: ./cmd/linkcheck --format json https://example.com
5. Test depth flag: ./cmd/linkcheck --depth 0 https://example.com

Check:
- Binary builds successfully
- Help text is clear
- Flags work correctly
- Output is properly formatted
- Exit codes are correct

Write results to .ai/verify_cli.md.
Write status.json: outcome=succeeded if all manual tests pass, outcome=failed with details otherwise."
    ]

    check_cli [shape=diamond, label="CLI OK?"]

    // Phase 10: Integration tests
    impl_integration [
        shape=box,
        timeout=1200,
        max_retries=2,
        prompt="Read .ai/spec.md section on testing.

Create integration tests in test/:
- integration_test.go: End-to-end tests
- Set up local HTTP test server with known structure
- Create test pages with working and broken links
- Create test robots.txt
- Run linkcheck against test server
- Verify correct broken links are detected
- Test depth limiting
- Test robots.txt compliance
- Test both output formats

Run: go test ./test/...

Write status.json: outcome=succeeded if integration tests pass, outcome=failed with failure_reason otherwise."
    ]

    verify_integration [
        shape=box,
        class="verify",
        timeout=300,
        goal_gate=true,
        prompt="Verify integration tests.

Run:
1. go test ./test/... -v
2. go test ./... (all tests)
3. go build ./...

Check:
- All integration tests pass
- All unit tests still pass
- Project builds cleanly

Write results to .ai/verify_integration.md.
Write status.json: outcome=succeeded if all pass, outcome=failed with details otherwise."
    ]

    check_integration [shape=diamond, label="Integration OK?"]

    // Phase 11: Final review
    review [
        shape=box,
        class="review",
        timeout=600,
        goal_gate=true,
        prompt="Read .ai/spec.md.

Perform final review of the complete linkcheck implementation:

1. Verify all requirements are met:
   - CLI takes URL as input
   - Crawls pages and finds all links
   - Checks each link for HTTP status
   - Reports broken links (4xx/5xx)
   - Respects robots.txt
   - Configurable crawl depth (default 1)
   - Outputs both human-readable and JSON formats

2. Code quality:
   - All packages have tests
   - Code is well-documented
   - Error handling is comprehensive
   - No obvious bugs or race conditions

3. Functionality:
   - Run: go build ./cmd/linkcheck
   - Test with real URLs
   - Verify robots.txt compliance
   - Verify depth limiting works
   - Verify both output formats work

4. Run complete test suite:
   - go test ./...
   - go vet ./...
   - go build ./...

Write detailed review to .ai/final_review.md.
Write status.json: outcome=succeeded if complete and working, outcome=failed with specific issues that need fixing."
    ]

    check_review [shape=diamond, label="Review OK?"]

    // Graph flow
    start -> expand_spec -> impl_setup -> verify_setup -> check_setup

    check_setup -> impl_types           [condition="outcome=succeeded"]
    check_setup -> impl_setup           [condition="outcome=failed", label="retry"]

    impl_types -> verify_types -> check_types
    check_types -> impl_robotstxt       [condition="outcome=succeeded"]
    check_types -> impl_types           [condition="outcome=failed", label="retry"]

    impl_robotstxt -> verify_robotstxt -> check_robotstxt
    check_robotstxt -> impl_checker     [condition="outcome=succeeded"]
    check_robotstxt -> impl_robotstxt   [condition="outcome=failed", label="retry"]

    impl_checker -> verify_checker -> check_checker
    check_checker -> impl_extractor     [condition="outcome=succeeded"]
    check_checker -> impl_checker       [condition="outcome=failed", label="retry"]

    impl_extractor -> verify_extractor -> check_extractor
    check_extractor -> impl_crawler     [condition="outcome=succeeded"]
    check_extractor -> impl_extractor   [condition="outcome=failed", label="retry"]

    impl_crawler -> verify_crawler -> check_crawler
    check_crawler -> impl_report        [condition="outcome=succeeded"]
    check_crawler -> impl_crawler       [condition="outcome=failed", label="retry"]

    impl_report -> verify_report -> check_report
    check_report -> impl_cli            [condition="outcome=succeeded"]
    check_report -> impl_report         [condition="outcome=failed", label="retry"]

    impl_cli -> verify_cli -> check_cli
    check_cli -> impl_integration       [condition="outcome=succeeded"]
    check_cli -> impl_cli               [condition="outcome=failed", label="retry"]

    impl_integration -> verify_integration -> check_integration
    check_integration -> review         [condition="outcome=succeeded"]
    check_integration -> impl_integration [condition="outcome=failed", label="retry"]

    review -> check_review
    check_review -> exit                [condition="outcome=succeeded"]
    check_review -> impl_cli            [condition="outcome=failed", label="fix"]
}