[stevenobiajulu]

Closes #218

What

• Gmail deleteMessage, drafts only. Calls users.drafts.delete, which gmail.compose authorizes, so there's no new scope and no restricted-scope verification. It accepts a draft id (r…) or the draft's message id; a message id is resolved through users.drafts.list, following pagination. The delete is always permanent (Gmail has no draft trash), whatever hard_delete says.
• Non-drafts refused. A sent or received message returns NOT_SUPPORTED: "Only drafts can be deleted with the current Gmail scopes … requires gmail.modify, which Agent Email does not request." No modify endpoint is called.
• Policy unchanged. Still needs AGENT_EMAIL_DELETE_ENABLED=true plus user_explicitly_requested_deletion: true. A Gmail draft delete doesn't need AGENT_EMAIL_HARD_DELETE_ENABLED: it's irreversible, but the draft was never sent. Passing hard_delete: true still requires the hard gate.
• IDs (acceptance item 3). update_draft, getDraft, and getDraftReplyStatus on Gmail also accept the draft's message id (the id list_emails(folder: "drafts") returns), using the same resolver. A non-draft returns NOT_A_DRAFT.
• delete_email errors are structured. Provider failures come back as NOT_SUPPORTED / MESSAGE_NOT_FOUND / DELETE_FAILED instead of being thrown.
• README, tool descriptions, and the OpenSpec change add-gmail-draft-delete are updated.

Out of scope: send_draft still takes only the draft id, because its duplicate-send guard keys on the id the caller passes, and accepting both ids would let the same draft get past the guard twice. The optional retire_draft helper is also left out.

Verification

• npm run build, npm run lint, npm run test:run (all green), check:spec-coverage PASS, openspec validate add-gmail-draft-delete --strict valid.
• New tests: 14 provider tests (mocked client) for draft id, message id, pagination, non-draft, unknown id, stale DRAFT label, the hard flag, error classification, an old client, the policy gate end to end through delete_email, and update_draft by message id. Also 2 googleapis-client tests and 3 core action error-mapping tests.
• Live smoke on a real Gmail mailbox (default scopes, built worktree CLI, throwaway self-addressed drafts only):
  • update_draft by message id: success, subject updated
  • delete_email by draft id: {"success":true}. Running it again: MESSAGE_NOT_FOUND
  • delete_email by message id: {"success":true}, and the draft no longer appears in list_emails(folder: drafts)
  • delete_email on an inbox message: NOT_SUPPORTED naming gmail.modify
  • With the gate unset: DELETE_DISABLED
  • The smoke found that Gmail gives a draft a new message id on every drafts.update, while the draft id stays stable. An old message id therefore gets MESSAGE_NOT_FOUND after an edit. This is now documented in the README and the delete_email id description.

[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 2026-10-18T03:02:45.206411Z	59f5870	PR opened

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.

[codecov]

Codecov Report

❌ Patch coverage is 96.55172% with 2 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
...ackages/provider-gmail/src/email-gmail-provider.ts	94.59%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 59f5870489

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]

Recognize not-found errors across package copies

When a provider resolves a different installed copy of @usejunior/email-core, its MessageNotFoundError fails this instanceof check and is returned as DELETE_FAILED instead of the promised MESSAGE_NOT_FOUND. This package-duplication scenario is explicitly handled for ProviderError by the structural isProviderError guard in providers/provider.ts, so the new not-found mapping needs an equivalent structural guard rather than relying solely on class identity.

Useful? React with 👍 / 👎.

[stevenobiajulu]

Peer review: Codex gpt-6.1-sol, low effort, read-only sandbox

Round 1: NEEDS-EXECUTION. The read-only sandbox blocked Vitest's temp files and GitHub access. I ran the blocked commands myself (provider-gmail 108 passed, email-core label 12 passed, both exit 0) and gave the reviewer a write-free way to run probes: node against the built dist/.

Round 2: changes requested. The reviewer backed its findings with executed node probes, including the real Gaxios retry machinery running on an in-memory adapter. All three are fixed in 78a9daf:

Severity	Finding	Fix
should-fix	A permanent draft delete under the soft gate contradicts email-security "Default to soft delete"	Added email-security MODIFIED Delete Policy delta: never-delivered items a provider cannot trash may be deleted under the soft gates; hard_delete: true still needs the hard gate; delivered mail keeps the soft default
should-fix	googleapis retries DELETE after a 503. If the first attempt deleted the draft, the retry's 404 was reported as MESSAGE_NOT_FOUND	A drafts.delete 404 after successful resolution now counts as success. Regression test added
nit	An old client without listDrafts let the raw 404 through as NOT_FOUND	Normalized to MessageNotFoundError. Test added

Probes passed for: stale and churned message ids failing closed, pagination, update_draft by message id, non-draft handling on every entry point, and no demonstrated wrong-draft deletion. A synthetic draft-id/message-id collision resolves to the draft namespace first.

Post-fix checks: build, lint, test:run (all green), spec coverage PASS, openspec validate --strict valid.

[stevenobiajulu]

✅ Post-merge smoke passed

Merged: 05481588ac04013220b441ea39c5d093da98ed96
Built from: main @ 0548158
Smoke: clean build → lint → npm run test:run → spec coverage → live Gmail end-to-end through the built CLI (email-agent-mcp call …, mailbox steven.obiajulu@gmail.com, default gmail.readonly + gmail.compose scopes, AGENT_EMAIL_DELETE_ENABLED=true, self-addressed throwaway drafts only)

Steps

• ✅ clean + build
• ✅ lint
• ✅ tests (620 + 237 + 145 + 225 + 31 passed)
• ✅ spec coverage: all scenarios have tests
• ✅ update_draft by message id: success, resolved to the stable draft id r-4810…, subject updated
• ✅ delete_email by message id: {"success":true}
• ✅ delete_email by draft id: {"success":true}. Repeating it gives MESSAGE_NOT_FOUND
• ✅ both smoke drafts are gone from list_emails(folder: "drafts") (0 left)
• ✅ delete_email on an inbox message: NOT_SUPPORTED naming gmail.modify; the message is still readable afterwards
• ✅ with AGENT_EMAIL_DELETE_ENABLED unset: DELETE_DISABLED

Notes, for full disclosure:

• The first smoke attempt hit one failure in security/watched-allowlist.test.ts > Hot reload on in-place write, an fs.watch timing test this PR doesn't touch. It passed 5/5 when run alone, and the full suite passed on rerun, so it looks like a flake under load.
• The second attempt failed in the smoke harness itself: zsh's echo mangled \n in JSON. It was fixed by switching to printf, and the two drafts that attempt left behind were deleted.

Visual evidence: n/a (no UI)
Prod: n/a. The repo doesn't deploy on merge, so locally installed MCP servers pick this up only after a release is tagged and the package is reinstalled.
Log: /tmp/automerge-smoke-219-1792293012.log (local)

Cleanup

• ✅ deleted remote branch feat/gmail-draft-delete
• ✅ removed worktree .worktrees/gmail-draft-delete
• ✅ deleted local branch (force delete after a squash merge; the PR is MERGED)