todo

The user invoked /procoder:todo with arguments:

The command below is the procoder binary on PATH.

Tasks live as Markdown files under .procoder/todo/, one file per task. This replaces throwaway checklists: every task carries a description that says what "done" looks like, acceptance criteria a reviewer could verify, and evidence of what proved them. procoder todo close is the quality controller — it REFUSES to close a task until every criterion is checked, the evidence is recorded, and the gate is clean. You do the work and the verification; the binary judges.

Subcommands (run the one matching the arguments; with no arguments, run list and report the open tasks):

• procoder todo add <title> — prints the task file and its path. Write the file, then REPLACE the placeholders: a real description (why the task exists and what done looks like), and one - [ ] line per testable acceptance criterion. Do this before starting the work — criteria written after the fact just describe whatever happened.

• procoder todo list — every task, open first.

• procoder todo show <id> — the full task file.

• procoder todo close <id> — the quality controller. Before running it: verify each criterion yourself (run the test, exercise the behaviour), check its box only when it is true, and fill ## Evidence with what you ran and what the output proved — one line per criterion. Then close. If it refuses, it names exactly what is missing: do that work, don't game the checkboxes. A checked box you did not verify is a lie the controller cannot catch — the evidence section exists so the user can.

  What counts as evidence — if you haven't run the verification command fresh, you cannot check the box:

  • "tests pass" = test output with 0 failures, from THIS state of the code — not an earlier run, not "should pass".
  • "bug fixed" = the test for the original symptom passes — and for a regression test, the red-green proof: revert the fix, watch the test FAIL, restore, watch it pass. A regression test never seen red proves nothing.
  • "build/lint clean" = the command's exit 0 and its output read, not extrapolated from a partial check.
  • a subagent's "done" = verified in the diff and by running its tests yourself — an agent's success report is a claim, not evidence.
  • "should", "probably", "seems to" in your evidence line means the box stays unchecked.

Rules:

• Never edit Status: by hand — only todo close moves a task to closed.
• Track every piece of multi-step work here, not in throwaway TodoWrite lists: these files survive the session and the user can read them.
• When a spec exists (procoder spec list), seed tasks from its acceptance criteria — one task per coherent group, criteria copied in.