Skip to content

Commands

Jump to bottom
AgenticMail edited this page May 18, 2026 · 1 revision

Commands

Every interaction with the bot starts with @agenticmail. The first mention on a line wins. The verb is the next token; everything after that is the args. Case-insensitive.

@agenticmail <verb> [args]

A bare @agenticmail with no verb defaults to summarize. An unknown verb (e.g. @agenticmail wibble) gets a help reply listing the valid commands.

Free plan

These verbs are pure read + AI-reply. No state-changing actions on your repo. Available to every install, no payment needed.

summarize

Posts a 2-paragraph summary of the thread. Works on both issues and pull requests. On PRs, the bot also pulls in up to 20 changed files (first 40 lines of each patch) so the summary reflects the actual diff, not just the PR description.

@agenticmail summarize

No args. Posted as a comment within ~5–10 seconds.

triage

Issue triage. The bot suggests labels (as a markdown list), a priority (low | medium | high | urgent), and notes whether the issue looks like a duplicate of anything in the thread. The bot does not apply labels or close the issue — triage is advisory.

@agenticmail triage

Auto-fires on every new issue too — no mention needed.

email <address>

Forwards the thread to a real inbox. The bot composes a short subject line + plain-text body and sends it to the address you specify. No embedded links to internal tools.

@agenticmail email engineering@acme.com

reply <prompt>

Drafts a follow-up comment from your prompt. Stays on-topic with the thread. Useful for "draft a polite ask for more info" or "explain the trade-off in two sentences" moments.

@agenticmail reply please confirm the migration plan by Friday
@agenticmail reply explain the trade-off in two sentences

handoff to <agent>

Re-routes the thread to another agent in your org. The bot acknowledges the handoff in a comment. Plumbing for inter-agent routing is in v2 — v1 just posts the acknowledgement and tags the receiving agent.

@agenticmail handoff to alice
@agenticmail handoff to billing-bot

The leading to is stripped from the args automatically.

link related

Looks for related open issues in the same repo. The bot can reference issue numbers it sees in the comment thread but won't invent numbers. If nothing is clearly related, it says so.

@agenticmail link related

Paid plan

These verbs perform real state-changing actions on your repo. Gated by plan check — invoking them from a Free-plan account returns a polite upgrade prompt with a Marketplace link, and the action never runs.

close [reason]

Closes the issue or PR. State reason maps as:

  • not planned, wontfix, won't fix, duplicate → state_reason: not_planned
  • everything else → state_reason: completed 
@agenticmail close
@agenticmail close not planned
@agenticmail close duplicate of #123

Requires Issues: write (covered by the default permission set).

merge [method]

Merges the pull request. Methods supported:

  • squash (default if no method given)
  • rebase
  • merge (or merge commit, true merge, no squash)
@agenticmail merge
@agenticmail merge rebase
@agenticmail merge merge commit

Requires Contents: write permission on the App. If GitHub's branch protection rules block the merge, the bot posts a ❌ comment with the status code and message. The bot won't auto-merge — there's no "auto-merge once CI passes" mode in v1.

review

Posts a formal Pull Request Review with event: COMMENT. The bot generates the review body via Claude with the diff in context (up to 20 files, 40 lines of patch each). Goes through POST /repos/.../pulls/N/reviews, not POST /comments — appears in the PR's review feed.

@agenticmail review

The bot never posts APPROVE or REQUEST_CHANGES. Human still has to approve for branch protection to be satisfied.

Failure modes

What happens What you'll see
Free-plan account invokes a paid verb 🔒 upgrade-required comment with Marketplace link
Rate limit exceeded (60/hour/installation) 🚦 cooldown comment with ETA
Action call fails (branch protection, missing perm) ❌ comment with HTTP status + GitHub error message
Unknown verb after @agenticmail 👋 help comment listing valid verbs
Bot mention but no valid verb (bare @agenticmail) Defaults to summarize
Bot mentions itself (its own comment fires issue_comment.created) Self-loop guard kicks in, returns in 0–1 ms

What gets sent to the LLM

Each call sends:

  • Repo name + issue/PR number + title
  • Original post body
  • Up to 20 most recent thread comments (oldest → newest)
  • On PRs: up to 20 changed files with diff stats and a truncated patch
  • The user who triggered the mention
  • The verb + any args

The system prompt instructs the model to:

  • Reply in Markdown, concise, no fabricated links or code
  • Always end with — [AgenticMail](https://agenticmail.io) · <verb>

Comment text leaves the function once, over TLS, to Anthropic's API. No persisted comment data on our end beyond the audit log (delivery ID, latency, token counts, no body text).

Clone this wiki locally