AgentKit
ReferenceCLI commands
BetaYou are reading docs for the beta channel (2.4.1-beta.1) — updated automatically on every beta release.Switch to stable →

CLI commands

ak watch dry-run

Preview what `ak watch start` would post without actually posting

Poll a GitHub repository for matching issues and simulate the full agent response pipeline - including rate-limit accounting - without posting any comment to GitHub. Reports how many issues would be responded to, skipped (already responded), or blocked by the rate ceiling.

Who it's for: Both: power devs validating filter configurations before enabling a live watch and non-technical users previewing automation impact.

When to use it: Before running ak watch start for the first time on a repo, to verify the label/author/keyword filters match the right issues and the rate limit is correctly configured. Safe to run anytime - nothing is posted.

What changes on disk: Read-only. No comments are posted to GitHub; no state file is updated.

Usage

ak watch dry-run <owner/repo> [flags]

Examples

ak watch dry-run owner/repo                        # preview all matched issues
ak watch dry-run owner/repo --label bug            # dry-run with label filter
ak watch dry-run owner/repo --max-per-hour 5       # preview with rate ceiling
ak watch dry-run owner/repo --json                 # machine-readable summary

Flags

FlagDescription
--agent stringAgent skill to run for each matched issue (default "engineer/scout")
--author stringsFilter to issues by these GitHub logins
-h, --helphelp for dry-run
--jsonEmit machine-readable JSON (implies --no-interactive)
--keyword stringsFilter to issues containing any of these keywords
--kits-dir stringOverride kits directory
--label stringsFilter to issues with any of these labels
--max-per-hour intRate limit ceiling (preview only) (default 10)
--no-interactiveDisable interactive prompts (CI-safe)
-q, --quietSuppress non-error output on stderr
-V, --verboseExtra diagnostic output on stderr (loses to --quiet)
-y, --yesAssume yes for all prompts

Output modes

ModeBehavior
prettydefault on TTY (colors, ASCII markers)
plainauto when stdout piped or --no-interactive
json--json (single-object envelope, NDJSON-safe)

Exit codes

CodeMeaning
0success
1runtime error
2invalid flags
3user-cancel (SIGINT, prompt-cancel)
  • ak watch — Watch a GitHub repository and auto-respond to issues

On this page