Cron-scheduled daemons in slug-registered directories never auto-fire (no run command found)

[technicalpickles]

I have two daemons configured in pitchfork. One is a long-running local web server, and the other is a separate command scheduled via cron to refresh its data every 15 minutes. The pitchfork.toml lives in a project directory and is registered via slug in ~/.config/pitchfork/config.toml. The web server stays running fine, but the cron-scheduled command never auto-fires.

After digging through the supervisor log, the cron itself is firing every 15 minutes exactly on schedule. It's just refusing to run the daemon:

2026-05-27 08:00:00 INFO  cron: triggering daemon my-project/my-fetch (retrigger: Finish)
2026-05-27 08:00:00 WARN  no run command found for cron daemon my-project/my-fetch
2026-05-27 08:15:00 INFO  cron: triggering daemon my-project/my-fetch (retrigger: Finish)
2026-05-27 08:15:00 WARN  no run command found for cron daemon my-project/my-fetch

This repeats every quarter-hour, indefinitely.

Setup

~/.config/pitchfork/config.toml:

[slugs.my-project]
dir = "/Users/me/path/to/my-project"
daemon = "my-serve"

/Users/me/path/to/my-project/pitchfork.toml:

namespace = "my-project"

[daemons.my-serve]
run = "./my-app serve"
depends = ["my-fetch"]

[daemons.my-fetch]
run = "./my-app fetch"
cron = { schedule = "0 */15 * * * *", retrigger = "finish" }

The supervisor is started via LaunchAgent on macOS, so its cwd is $HOME (or /, depending on how launchd resolves it). The project directory is not in that cwd's ancestry.

What's happening

Tracing through the source on main:

• cron_watch in src/supervisor/watchers.rs calls self.get_daemon_run_command(&id).
• That function (in src/supervisor/retry.rs) calls PitchforkToml::all_merged(), which delegates to PitchforkToml::all_merged_from(&env::CWD).
• list_paths_from(cwd) discovers project configs by walking UP from cwd via xx::file::find_up_all. It does not consult slugs.
• Because the supervisor's cwd is $HOME, the project's pitchfork.toml at ~/path/to/my-project/pitchfork.toml is never loaded.
• pt.daemons.get(id) returns None, the WARN fires, and the daemon is skipped.

Meanwhile, pitchfork restart my-fetch works fine because the restart command resolves the run command via the slug, not via CWD discovery. So manual restarts and depends-bootstraps fire the daemon as expected, but the cron schedule cannot.

Reproduction

1. Register a daemon via slug in ~/.config/pitchfork/config.toml, with the actual daemon config (including a cron schedule) in the project's local pitchfork.toml.
2. Start the supervisor via LaunchAgent (not from inside the project directory).
3. Watch the supervisor log. Every cron tick logs INFO cron: triggering daemon ... followed by WARN no run command found for cron daemon .... The daemon never runs.
4. pitchfork restart <daemon> works correctly. The cron just doesn't.

Suggested fix

get_daemon_run_command should look up the daemon's source pitchfork.toml via the slug registry rather than walking up from env::CWD. The supervisor already knows the slug's dir (it's in ~/.config/pitchfork/config.toml), so the run command could be resolved from PitchforkToml::all_merged_from(slug_dir) instead.

Workaround

pitchfork restart <daemon> whenever the data feels stale. Setting up a system launchd timer also works as a fallback.

Environment

• pitchfork 2.11.0
• macOS (Darwin 25.5.0), LaunchAgent supervisor
• mise-managed install

Happy to test a fix or pull more logs if helpful.

[gaojunran]

Get! Will fix it soon