track

track skill

This skill is the safe-invocation wrapper around bin/track.sh. Its sole job is to translate one of the /track:* slash commands into a Bash tool invocation that runs the dispatcher correctly. All ledger logic, locking, atomic writes, git commits, jq math, and error messages live in bin/track.sh — do not reimplement any of that here.

Locating the dispatcher

The dispatcher lives at ${CLAUDE_PLUGIN_ROOT}/bin/track.sh. If $CLAUDE_PLUGIN_ROOT is unset, fall back to the absolute path inferred from the SKILL.md location: <plugin root>/bin/track.sh where the plugin root is the parent of skills/track/. Either path resolves to the same script.

Invocation patterns

For each subcommand, run the Bash tool with the exact command shape below. Do NOT add extra flags, do NOT pre-process arguments, do NOT interpret the output — print whatever the dispatcher prints, verbatim.

/track:start

User invocation: /track:start <client> [--at HH:MM | --ago <duration>] (with $ARGUMENTS containing the client name and optional backdating flag)

Bash invocation:

bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" start $ARGUMENTS

The <client> argument is constrained by the dispatcher to alphanumeric + dash + underscore characters. The optional flags are also dispatcher-validated (--at requires HH:MM, --ago requires <Nh><Nm><Ns> like 30m, 1h30m, 90s). Both are safe to pass positionally.

Backdating flag-parsing rule: if $ARGUMENTS contains --at <VAL> or --ago <VAL> tokens, they are flag-value pairs consumed by the dispatcher. They are only meaningful at the start of $ARGUMENTS, before the client name token. Order between --at/--ago and the client is flexible (either start vytl --ago 30m or start --ago 30m vytl works), but pass through $ARGUMENTS as the user typed it — let the dispatcher's flag loop handle the parsing.

/track:stop

User invocation: /track:stop [--at HH:MM | --ago <duration>] [note...] (with $ARGUMENTS containing optional backdating flag(s) followed by an optional note)

Critical: the note is piped through stdin, never passed as an argument. This is a security boundary — the note is the only user-controlled input string in the system, and stdin-piping eliminates any possibility of shell metacharacter interpretation. Backdating flags go on argv (the dispatcher rejects any positional arg as ambiguous).

Flag-parsing rule (getopt convention): when splitting $ARGUMENTS, the dispatcher consumes --at <VAL> or --ago <VAL> only when they appear at the start of $ARGUMENTS, before the first non-flag token. After the first non-flag token, all remaining tokens are part of the note — including any literal --at or --ago strings inside it. This prevents a note like "fixed bug --at line 47" from being mis-parsed as a flag.

So when invoking:

• If $ARGUMENTS is empty (no flags, no note):

  bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" stop
  

• If $ARGUMENTS starts with --at <VAL> or --ago <VAL> (with no note after):

  bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" stop <FLAG> <VAL>
  

• If $ARGUMENTS contains only a note (no leading flags):

  printf '%s' "<NOTE>" | bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" stop
  

• If $ARGUMENTS starts with a flag AND has a note after:

  printf '%s' "<NOTE>" | bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" stop <FLAG> <VAL>
  

In every case <NOTE> is the literal text of the note portion (after any leading flag tokens are consumed), quoted as a shell string. Use printf '%s' rather than echo to avoid escape-sequence interpretation of the note content.

The dispatcher's cmd_stop parses argv flags first, then checks [ ! -t 0 ] and reads stdin for the note when present. It refuses end-before-start (the only validation on backdating), preserving the active timer for retry.

If the timer is currently paused and the user did not pass --at/--ago, the dispatcher uses the pause point as the end automatically — the user does not need to resume before stopping.

/track:pause

User invocation: /track:pause [--at HH:MM | --ago <duration>] (with $ARGUMENTS containing the optional backdating flag)

Bash invocation:

bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" pause $ARGUMENTS

/track:pause takes no positional args (no client, no note). --at/--ago backdate the moment the pause began — useful if the user forgot to pause when they actually stepped away. The dispatcher refuses a pause timestamp before the timer's start, and refuses to pause if the timer is already paused or if no timer is running.

/track:resume

User invocation: /track:resume [--at HH:MM | --ago <duration>] (with $ARGUMENTS containing the optional backdating flag)

Bash invocation:

bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" resume $ARGUMENTS

/track:resume takes no positional args. --at/--ago backdate the resume moment. The dispatcher refuses a resume timestamp before the pause point, and refuses to resume if no timer is running or the timer isn't currently paused.

/track:status

User invocation: /track:status (no arguments)

Bash invocation:

bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" status

/track:report

User invocation: /track:report [--week | --month YYYY-MM] [--client <name>] (with $ARGUMENTS containing the optional flags)

Bash invocation:

bash "${CLAUDE_PLUGIN_ROOT}/bin/track.sh" report $ARGUMENTS

Flag values are constrained by the dispatcher (regex check on --month YYYY-MM and the client name) and are safe to pass positionally.

Output handling

Print the dispatcher's stdout and stderr verbatim. Do not summarize, do not reformat, do not add commentary. The dispatcher's output is designed to be the user-facing surface; wrapping it in chat prose adds noise.

If the dispatcher exits non-zero, surface that to the user — the error message is already user-facing and includes recovery instructions where applicable.