[tracking202]

Backlog item mcp-setup-objects-crud.

The MCP connector (p202 mcp and the hosted server) could only list and create traffic sources (ppc_network), traffic source accounts (ppc_account) and categories (aff_network). So when asked to rename "Google Ads - Main" to "Main" it could only make a second source, and an account's default flag could not be changed after it was created. This PR adds:

• ppc_network_update, ppc_network_delete
• ppc_account_update, ppc_account_delete
• aff_network_update, aff_network_delete

These are the same kind of tool as landing_page_update/_delete. Each runs the p202 <entity> update|delete command, and those commands, like PUT/DELETE /api/v3/{ppc-networks,ppc-accounts,aff-networks}/{id}, already existed. So nothing changes in the API or in the CLI commands.

What happens on a delete

• The delete follows the existing convention: a soft delete (the *_deleted flag), run through the tool with --force once the call is approved, and dry_run: true gives a preview first (deleteNote).
• Children are not touched. This matches Setup › Traffic Sources / Setup › Categories (ppc_accounts.php, aff_networks.php set the flag on that one row) and the API's Controller::deleteRecord(). A campaign delete likewise leaves its trackers. A deleted source's accounts and a deleted category's campaigns stay and still name their parent. The dry run's cascade is []. I did not add a cascade or a refusal, because either would make the API disagree with the Setup pages. Instead, the delete tools' descriptions say what stays and name the tool that removes it (ppc_account_delete, campaign_delete). The OpenAPI DELETE descriptions and the CLI docs say the same.

Account scoping (IDOR)

• Writes are scoped by the base controller: get()/update()/deleteRecord() filter on user_id, and assertLinksOwned checks ppc_network_id on an account update.
• New tests/Api/V3/SetupObjectsWriteScopeIntegrationTest:
  • Another account's source, account or category answers 404 for an update, a delete and a dry run, and its row stays byte-for-byte the same.
  • A rename keeps the id and the row count, and the children still name the parent.
  • A delete soft-deletes that one row only.
• Hosted scope: AiConnect::WRITE_AREAS already holds ppc-networks, ppc-accounts and aff-networks. Existing connections need no reconnect, and neither AiConnect.php nor connect-ai.php changes. Its copy already says "change and delete ... traffic sources and categories".

Also in this PR

• p202 search tasks for the three update commands.
• Agent eval setup-source-001: rename in place, no create or delete. Reference agent branch included.
• Docs: documentation/cli/10-go-cli.md, docs/cli.md, docs/cli-agent.md, docs/openapi.yaml.
• Changelogs: changelogs.txt and 202-config/changelogs.json.

Overlap with #260, #265, #267

• The shared files have small additions, placed away from where 1.9.79: text_ad_create, text_ad_update and text_ad_delete in the MCP connector #260 and 1.9.79: redirector create, update, delete and whole-rule-set writes in the MCP connector #265 insert:
  • the catalogue entries go after each create;
  • the hosted inputs go after landing_page_update;
  • the changelog entry is first in the 1.9.79 list, while the others append at the end;
  • the reference agent branch sits just before *[Dd]elete*).
• One line will conflict for sure: TestMCPCatalogStaysInBudget's budget. This PR sets it to 85000 and 1.9.79: text_ad_create, text_ad_update and text_ad_delete in the MCP connector #260 to 84000. Whichever merges second should add the two increases together (about 4,600 here and 3,400 there).
• crud.go, AiConnect.php, connect-ai.php, the legacy Setup pages and api/v1/functions.php are not touched.

Verification (local, macOS, PHP 8.5, MySQL 8.0 in Docker)

• cd go-cli && go vet ./cmd/ && go test ./cmd/ passed. The TestMCPToolArgv cases are new. The hosted-scope test runs every tool in each mode against AiConnect's areas.
• SetupObjectsWriteScopeIntegrationTest: 3 tests, 84 assertions, OK. It is not vacuous: with the user_id scope removed from PpcNetworksController, the cross-account test fails.
• OwnedReferencesIntegrationTest, PayloadValidationIntegrationTest and IntegrationSuiteSelectionTest passed.
• verify.sh --changed --since origin/1.9.79:

  go PASS, golangci PASS, openapi PASS, patterns PASS, phpcs PASS, syntax PASS
  phpstan FAIL  phpstan.neon.dist passed, phpstan-strict.neon failed
  unit FAIL     local PHP is 8.5, CI pins 8.3
  

  The phpstan-strict findings are in Connection.php and LookupRepositoryFactory.php, which this PR does not touch. The unit tier's failures are the same 14 test files with identical counts on origin/1.9.79 (checked with these changes stashed), so they come from PHP 8.5, not from this change.
• Not run locally: a live instance or the eval case end to end. The Agent Evals job on this PR is the first run of setup-source-001.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-27T09:36:31.897711Z	62d30b1	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

Code review: CLEAN

For da4d5df: 0 issue(s). The review is encrypted for the maintainers in the review-findings artifact of this run.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: da4d5dfce2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Restrict none to report filters

The new ppc_account_update tool inherits mcpIDFlagPattern for its ppc_network_id flag, whose schema accepts none even though that exception is intended only for report filters. Thus an MCP client can submit a schema-valid request such as {"id":"9","ppc_network_id":"none"}, but the PPC account controller requires an integer and rejects the update with a 422. Generate the numeric-only pattern for this update field, or make the none exception depend on the command being a report.

Useful? React with 👍 / 👎.

[tracking202]

Fixed in 62d30b1: mcpIDFlagPattern now takes the tool kind and offers none for ppc_network_id only on read tools (the report filters). On write tools (ppc_account_create, ppc_account_update) the field is digits only, and {"ppc_network_id":"none"} is refused before any request. TestMCPToolArgv pins it: none is refused on ppc_account_update and ppc_account_create, and report_summary still takes none. The two refusal cases fail against the previous mcp_catalog.go. go test ./cmd/ passes.

[github-actions]

Adversarial review: NON-BLOCKING

For 62d30b1: 3 finding(s). The report is encrypted for the maintainers in the review-findings artifact of this run.

Rewritten on every push; the check fails on BLOCKING.