osdu - AgentSkills CN | 技能图谱

Name: osdu
Rating: 78
Author: danielscholl

<osdu-command> <objective> Analyze GitLab CI/CD test job reliability for OSDU platform projects, tracking test job pass/fail status across pipeline runs to identify flaky tests and provide quality metrics. </objective> <triggers> <use-when> <trigger>OSDU project test status queries ("how is {project} looking", "partition test quality")</trigger> <trigger>Flaky test detection ("are there flaky tests in {project}")</trigger> <trigger>Pipeline health monitoring ("recent pipeline failures")</trigger> <trigger>Cloud provider comparison ("azure vs aws test reliability")</trigger> <trigger>Stage-specific analysis ("unit test status", "integration test failures")</trigger> </use-when>

code

<skip-when>
  <condition>Individual test case tracking (we track job-level, not test-level)</condition>
  <condition>Non-test jobs (build, deploy, lint, security scans)</condition>
  <condition>Non-OSDU projects or non-GitLab CI systems</condition>
  <condition>Real-time monitoring (data is from completed pipelines only)</condition>
</skip-when>

</triggers> <tracking-scope> <hierarchy> Pipeline Run → Test Stage (unit/integration/acceptance) → Test Job → Test Suite (many tests) </hierarchy>

code

<capabilities>
  <supported>Test job pass/fail status across multiple pipeline runs</supported>
  <supported>Flaky test job detection (jobs that intermittently fail)</supported>
  <supported>Stage-level metrics (unit/integration/acceptance)</supported>
  <supported>Cloud provider breakdown (azure, aws, gcp, ibm, cimpl)</supported>
  <unsupported>Individual test results (not tracked)</unsupported>
  <unsupported>Non-test jobs like build, deploy, lint</unsupported>
</capabilities>

<example>
  Pipeline #1: job "unit-tests-azure" → PASS (100/100 tests passed)
  Pipeline #2: job "unit-tests-azure" → FAIL (99/100 tests passed)
  Pipeline #3: job "unit-tests-azure" → PASS (100/100 tests passed)
  Result: This job is FLAKY (unreliable across runs)
</example>

</tracking-scope> <query-strategy importance="critical"> <token-usage-reference> <query type="status.py --format json" pipelines="5" projects="1" tokens="900" speed="10s" safety="always-safe"/> <query type="analyze.py --format json" pipelines="5" projects="1" tokens="35000" speed="30s" safety="use-cautiously"/> <query type="analyze.py --format json" pipelines="10" projects="1" tokens="68000" speed="60s" safety="heavy"/> <query type="analyze.py" pipelines="10+" projects="multiple" tokens="100000+" speed="120s+" safety="avoid"/> <query type="analyze.py" pipelines="all" projects="30" tokens="200000+" speed="180s+" safety="exceeds-limit"/> </token-usage-reference>

code

<progressive-approach mandatory="true">
  <step number="1" name="start-light">
    <action>Use status.py for quick overview</action>
    <command>script_run osdu status.py --format json --pipelines 3 --project {name}</command>
    <rationale>Lightweight, fast, safe token usage</rationale>
  </step>

  <step number="2" name="deep-dive" condition="only-if-needed">
    <action>Use analyze.py with strict filters</action>
    <command>script_run osdu analyze.py --format json --pipelines 5 --project {name} --stage unit</command>
    <rationale>Heavy query, use only when status insufficient</rationale>
  </step>

  <step number="3" name="never-query-all">
    <action>ALWAYS specify --project to avoid 30-project scan</action>
    <rationale>Prevents token limit exceeded error</rationale>
  </step>
</progressive-approach>

<format-selection>
  <format type="json">
    <use-when>Extracting specific metrics or calculating statistics</use-when>
    <use-when>Building summaries or comparisons</use-when>
    <use-when>Parsing structured data programmatically</use-when>
    <use-when importance="critical">ALWAYS for status.py (lightweight, parseable)</use-when>
  </format>

  <format type="markdown">
    <use-when>Analyze.py queries (10x smaller than JSON, still readable)</use-when>
    <use-when>Creating reports for sharing</use-when>
    <use-when>Need human-readable tables without parsing</use-when>
    <use-when>Token budget is tight</use-when>
  </format>

  <format type="terminal" status="never-use">
    <avoid-because>Includes ANSI codes and colors, hard to parse</avoid-because>
    <avoid-because>Only for direct human terminal viewing</avoid-because>
  </format>
</format-selection>

</query-strategy> <available-projects count="30"> <core-services description="Most Common Queries"> <project name="partition" description="Multi-tenant data partitioning"/> <project name="storage" description="Blob/file storage service"/> <project name="indexer-service" description="Search indexing"/> <project name="search-service" description="Search API"/> <project name="entitlements" description="Auth/permissions"/> <project name="legal" description="Legal tags/compliance"/> <project name="schema-service" description="Schema registry"/> <project name="file" description="File metadata management"/> </core-services>

code

<domain-services>
  <project name="wellbore-domain-services" description="Wellbore data"/>
  <project name="well-delivery" description="Well delivery workflows"/>
  <project name="seismic-store-service" description="Seismic data storage"/>
  <project name="dataset" description="Dataset management"/>
  <project name="register" description="Data registration"/>
  <project name="unit-service" description="Unit conversion"/>
</domain-services>

<reference-services>
  <project name="crs-catalog-service" description="Coordinate reference systems"/>
  <project name="crs-conversion-service" description="CRS conversion"/>
</reference-services>

<ddms-services>
  <project name="rafs-ddms-services" description="R&D data management"/>
  <project name="eds-dms" description="Engineering data management"/>
</ddms-services>

<workflow-processing>
  <project name="ingestion-workflow" description="Data ingestion pipelines"/>
  <project name="indexer-queue" description="Indexing queue management"/>
  <project name="notification" description="Event notifications"/>
  <project name="segy-to-mdio-conversion-dag" description="Seismic format conversion"/>
</workflow-processing>

<infrastructure>
  <project name="infra-azure-provisioning" description="Azure infra provisioning"/>
  <project name="os-core-common" description="Shared core libraries"/>
  <project name="os-core-lib-azure" description="Azure-specific libs"/>
</infrastructure>

<other-services>
  <project name="geospatial" description="Geospatial services"/>
  <project name="policy" description="Policy engine"/>
  <project name="secret" description="Secret management"/>
  <project name="open-etp-client" description="ETP protocol client"/>
  <project name="schema-upgrade" description="Schema migration tools"/>
</other-services>

<cloud-providers>
  <provider code="azure" name="Microsoft Azure"/>
  <provider code="aws" name="Amazon Web Services"/>
  <provider code="gcp" name="Google Cloud Platform"/>
  <provider code="ibm" name="IBM Cloud"/>
  <provider code="cimpl" name="CIMPL (Venus) provider"/>
</cloud-providers>

</available-projects> <prerequisites> <requirement>osdu-quality CLI installed: uv tool install git+https://community.opengroup.org/danielscholl/osdu-quality.git</requirement> <requirement>GitLab authentication (choose one): - GITLAB_TOKEN environment variable, OR - glab CLI authenticated (glab auth login) </requirement> <requirement>Access to OSDU GitLab projects</requirement> </prerequisites> <scripts> <script name="status.py" recommendation="preferred"> <purpose>Quick overview of latest pipeline test results by stage</purpose>

code

  <when-to-use>
    <scenario>Initial health check ("how is {project} doing?")</scenario>
    <scenario>Recent pipeline status</scenario>
    <scenario>Quick pass/fail overview</scenario>
    <scenario importance="high">Default choice for most queries</scenario>
  </when-to-use>

  <token-impact>~900 tokens per project (very safe)</token-impact>

  <options>
    <option name="--pipelines N" default="10" recommended="3-5">Analyze last N pipelines</option>
    <option name="--project NAME" required="true">Specify project (see list above)</option>
    <option name="--format json" required="true">Structured output for parsing</option>
    <option name="--venus">Filter to CIMPL (Venus) provider pipelines only</option>
    <option name="--no-release">Exclude release tag pipelines (master/main branch only)</option>
  </options>

  <examples>
    <example description="Quick status check (recommended starting point)">
      script_run osdu status.py --format json --pipelines 3 --project partition
    </example>
    <example description="Check specific project without releases">
      script_run osdu status.py --format json --pipelines 5 --project storage --no-release
    </example>
    <example description="Venus provider status">
      script_run osdu status.py --format json --pipelines 3 --project indexer-service --venus
    </example>
  </examples>
</script>

<script name="analyze.py" recommendation="use-cautiously">
  <purpose>In-depth flaky test detection and reliability metrics across many pipeline runs</purpose>

  <when-to-use>
    <scenario>After status.py shows issues</scenario>
    <scenario>Flaky test job detection needed</scenario>
    <scenario>Calculating pass rates over time</scenario>
    <scenario>Provider comparison analysis</scenario>
    <scenario importance="critical">Only with strict filters (project + stage or provider)</scenario>
  </when-to-use>

  <token-impact>
    <impact pipelines="5" projects="1">~35K tokens (moderate)</impact>
    <impact pipelines="10" projects="1">~68K tokens (heavy)</impact>
    <impact projects="multiple">Can exceed 200K token limit ❌</impact>
  </token-impact>

  <critical-rules>
    <rule priority="1">ALWAYS specify --project (never scan all 30 projects)</rule>
    <rule priority="2">Start with --pipelines 5 (not default 10)</rule>
    <rule priority="3">Add --stage or --provider for additional filtering</rule>
    <rule priority="4">Use --format markdown if token budget is tight (10x smaller than JSON)</rule>
    <rule priority="5">Only use if status.py insufficient</rule>
  </critical-rules>

  <options>
    <option name="--pipelines N" default="10" recommended="5">Analyze last N pipelines</option>
    <option name="--project NAME" required="true">Specific project (comma-separated for multiple)</option>
    <option name="--format FORMAT" required="true" recommended="markdown">Use markdown to save tokens</option>
    <option name="--stage STAGE">Filter by test stage (unit/integration/acceptance)</option>
    <option name="--provider PROVIDER">Filter by cloud provider (azure/aws/gcp/ibm/cimpl)</option>
  </options>

  <examples>
    <example description="Analyze flaky tests (safe query)">
      script_run osdu analyze.py --format markdown --pipelines 5 --project partition --stage unit
    </example>
    <example description="Provider comparison (focused)">
      script_run osdu analyze.py --format markdown --pipelines 5 --project storage --provider azure
    </example>
    <example description="Multi-project with strict filter (use cautiously)">
      script_run osdu analyze.py --format markdown --pipelines 5 --project partition,storage --stage unit
    </example>
  </examples>
</script>

</scripts> <query-patterns> <pattern name="quick-health-check"> <description>Best approach: Start with status.py</description> <command>script_run osdu status.py --format json --pipelines 3 --project partition</command> </pattern>

code

<pattern name="flaky-test-detection">
  <step number="1">Check status</step>
  <command>script_run osdu status.py --format json --pipelines 5 --project partition</command>
  <step number="2">If issues found, deep dive with analyze.py</step>
  <command>script_run osdu analyze.py --format markdown --pipelines 5 --project partition --stage unit</command>
</pattern>

<pattern name="provider-comparison">
  <description>Compare Azure vs AWS for specific project/stage</description>
  <command>script_run osdu analyze.py --format markdown --pipelines 5 --project storage --stage integration --provider azure</command>
  <command>script_run osdu analyze.py --format markdown --pipelines 5 --project storage --stage integration --provider aws</command>
</pattern>

<pattern name="stage-specific-analysis">
  <description>Focus on unit tests only</description>
  <command>script_run osdu analyze.py --format markdown --pipelines 5 --project entitlements --stage unit</command>
</pattern>

</query-patterns> <anti-patterns importance="critical"> <dont-do description="Query all projects without filters"> <bad-example>script_run osdu analyze.py --format json --pipelines 10</bad-example> <reason>Will exceed 200K token limit!</reason> </dont-do>

code

<dont-do description="Use high pipeline counts without project filter">
  <bad-example>script_run osdu analyze.py --format json --pipelines 20</bad-example>
  <reason>Takes 3+ minutes, huge output</reason>
</dont-do>

<dont-do description="Use terminal format in agent context">
  <bad-example>script_run osdu status.py --format terminal --project partition</bad-example>
  <reason>Includes ANSI codes, hard to parse</reason>
</dont-do>

<dont-do description="Jump straight to analyze.py">
  <bad-example>script_run osdu analyze.py --format json --pipelines 10 --project partition</bad-example>
  <reason>Heavy query when status.py would suffice</reason>
</dont-do>

</anti-patterns> <best-practices> <practice priority="1" name="Progressive Disclosure"> Always start with status.py, only use analyze.py if needed </practice> <practice priority="2" name="Explicit Formats"> Always specify --format json or --format markdown </practice> <practice priority="3" name="Project Specificity"> Always include --project {name} to avoid all-30-projects scan </practice> <practice priority="4" name="Conservative Pipelines"> Start with --pipelines 3-5, increase only if necessary </practice> <practice priority="5" name="Add Filters"> Use --stage or --provider to narrow scope </practice> <practice priority="6" name="Markdown for Heavy"> Prefer --format markdown for analyze.py (10x token savings) </practice> </best-practices> <output-formats> <format type="json" size="large" agent-usage="status.py-only"> <best-for>Structured parsing, metrics extraction</best-for> </format> <format type="markdown" size="medium" agent-usage="analyze.py-preferred"> <best-for>Reports, sharing, analyze.py queries</best-for> </format> <format type="terminal" size="small" agent-usage="never"> <best-for>Human viewing in terminal with colors</best-for> </format> </output-formats> <error-handling> <error condition="osdu-quality CLI not installed">Clear message with installation command</error> <error condition="GITLAB_TOKEN not set">Message about authentication requirements</error> <error condition="GitLab API errors">API error details</error> <error condition="Invalid project/filter">List of valid options</error> <error condition="Query timeout">Suggestion to reduce --pipelines or add filters</error> </error-handling> <instructions> <guideline priority="critical">Always follow progressive query approach</guideline> <guideline priority="critical">Never query without --project filter</guideline> <guideline>Start with minimal pipeline counts</guideline> <guideline>Use markdown format for analyze.py to save tokens</guideline> <guideline>Apply stage or provider filters when possible</guideline> </instructions> </osdu-command>