Update Dotty Docs

Orchestration model

You orchestrate: find the PR, drive CI-failure decisions, review, create the docs PR. Delegate context-heavy work to subagents so bulk output (diffs, ~90 CI-check lines, huge job logs, the docs .mdx search) stays out of your context - you get only the slim result. Never run gh pr diff, gh pr checks, gh api .../jobs/<id>/logs, or the docs .mdx search in the main thread.

Every subagent prompt MUST tell the subagent to:

• Run its gh/git/file commands directly.
• NOT invoke any skill and NOT spawn further subagents.
• Return ONLY the compact schema specified - not raw command output.

Step 0: Prerequisites, then sync the docs fork

1. Run gh auth status; if it fails, report the error and stop (covers gh-missing and not-authenticated). Git is assumed.
2. Capture the user's GitHub login: gh api user --jq '.login' -> $GH_USER. Verify the fork exists: gh api repos/$GH_USER/docs-website --jq '.full_name'. If no fork, stop (do not create one).
3. Sync the fork's develop to newrelic/docs-website:develop before anything else. The fork is often hundreds of commits behind, and a stale base makes the later PR diff misleading:

   gh api --method POST repos/$GH_USER/docs-website/merge-upstream -f branch=develop
   

   Read merge_type in the response: fast-forward means it synced, none with "not behind the upstream" means it was already current. Both are success. Report which one to the user. Any other result - stop and report.

This API call needs no clone, so it runs before Step 5 clones the fork. Do not sync with git reset --hard: settings deny it, and it needs a clone anyway.

Step 1: Find the open Dotty PR

Run yourself (you need the PR number in Step 8):

gh pr list --repo newrelic/newrelic-dotnet-agent --search "dotty" --state open --limit 1 --json number,title,body,url

If none found, stop. Otherwise record the PR number.

Step 2 & 3: Intake diff + CI status (dispatch ONE subagent)

Dispatch one read-only subagent (Explore). Prompt (substitute the PR number):

Run these two commands directly (do not invoke any skill; do not spawn subagents):

• gh pr diff <number> --repo newrelic/newrelic-dotnet-agent
• gh pr checks <number> --repo newrelic/newrelic-dotnet-agent

From the diff, for each changed <PackageReference> where the Version changed, extract: package name (Include), old version, new version, and section from the Condition:

• net4xx -> Framework
• net8.0/net10.0/etc. -> Core
• no Condition -> both Collapse identical package+old+new rows that differ only by TFM into one row with the combined section. Ignore NewRelic.Agent.Api bumps.

From the checks, look only at jobs matching Run IntegrationTests (*), Run Unbounded Tests (*), and Run Linux Container Tests (*). Report the overall verdict.

Return ONLY this, nothing else:

1. A markdown table: package | old | new | section.
2. CI: either the literal all passing, or a table of failing jobs: name | job_id | url (extract job_id from the job URL's trailing number).

If CI is all passing, skip to Step 4. Otherwise, for each failing job use AskUserQuestion with three options:

• Analyze - see Step 3a (dispatch a log-analysis subagent).
• Skip - exclude any package that job could implicate from the docs update; continue.
• Stop - end the skill.

After handling all failing jobs, summarize which packages are safe to update.

Step 3a: Analyze a failing job (dispatch ONE subagent per job, in parallel)

On Analyze, dispatch one read-only subagent (Explore) per failing job, all in a single message so they run concurrently. Prompt each (substitute job_id and paste the package-change table):

Run this directly (do not invoke any skill; do not spawn subagents): gh api repos/newrelic/newrelic-dotnet-agent/actions/jobs/<job_id>/logs

The following packages were bumped in this PR:

Correlate the failing errors in the log with these packages. Check for transitive dependency conflicts (example: OpenAI 2.9.0 removed a type that Azure.AI.OpenAI depends on - inspect dependency chains in tests/Agent/IntegrationTests/SharedApplications/Common/MFALatestPackages/MFALatestPackages.csproj in the current repo if needed).

Return ONLY: a short list of implicated package names, each with a one-line reason. If nothing is implicated, return no packages implicated.

Aggregate the returned lists. Present implicated packages to the user and confirm which are safe to update before continuing.

Step 4: Flag major version bumps

From the slim table (Step 2 & 3), for any major version change (e.g., 3.x -> 4.x), report it prominently - it may block merging the Dotty PR.

Step 5: Prepare the docs repo

Step 0 already synced the fork, so a plain clone of develop gives a current base. Clone to a known path (not a bash tempdir - the path must survive across tool calls). Use C:\tmp\docs-website-dotty-<date> or similar; record the absolute path for later steps.

git clone --branch develop https://github.com/$GH_USER/docs-website.git <path>
cd <path>
git checkout -b dotnet/dotty-updates-YYYY-MM-DD

All subsequent steps run inside the clone path.

Step 6: Update the docs file (dispatch ONE subagent)

Dispatch one edit-capable subagent (general-purpose). Prompt (substitute the clone path and paste the safe-to-update package table):

Work inside <clone-path>. Do not invoke any skill; do not spawn subagents. Edit this file only: src/content/docs/apm/agents/net-agent/getting-started/net-agent-compatibility-requirements.mdx

It has two tabbed sections: .NET Core (id="core") and .NET Framework (id="framework"). For each package below, update the version in the section(s) matching its section column (both = update both tabs).

Packages to update:

Package-to-docs mapping: not every package has a docs entry - skip silently if the name is not found. Always search the file for the package name. Non-obvious mappings: Microsoft.Azure.Cosmos -> Cosmos DB, CouchbaseNetClient -> Couchbase, EnyimMemcachedCore -> Memcached, AWSSDK.BedrockRuntime -> AWS Bedrock, Elastic.Clients.Elasticsearch -> Elasticsearch, MySql.Data -> MySQL, Npgsql -> PostgreSQL, AWSSDK.DynamoDBv2 -> DynamoDB. Logging/LLM/AWS-SDK packages typically map by exact name.

Version formats - change only the version number, nothing else:

• Bullet (Datastores, Messaging): * Latest verified compatible version: 3.57.0
• Table column (LLM, Logging - last <td> in row): the version inside the <td>.

Return ONLY: a table package | old | new | section | updated? where updated? is yes or no docs entry.

Step 7: Present for review

Present, from inside the clone path:

1. The Step 6 summary table (package, old, new, section, updated?).
2. Major version bumps from Step 4.
3. The full diff - run git diff yourself (small, controlled change).

Do not commit or push without explicit approval.

Step 8: After approval - commit, push, PR, comment

Never use "Dotty" (team-specific term) in the commit message or PR title/body; use "latest supported library versions" phrasing. Branch name may keep dotty.

1. Commit: chore(.net agent): Update compatibility docs for latest supported library versions
2. Push the branch to origin.
3. Create PR against upstream, ready for review (no --draft):

   gh pr create --repo newrelic/docs-website --base develop --head $GH_USER:<branch> \
     --title "chore(.net agent): Update compatibility docs for latest supported library versions" \
     --body "<brief list of packages updated>"
   

4. Comment on the Dotty PR:

   gh pr comment <dotty-pr-number> --repo newrelic/newrelic-dotnet-agent \
     --body "Compatibility docs update PR created: <docs-pr-url>"
   

5. Remove the clone directory (use the absolute path recorded in Step 5).