Repository navigation
Description
Found during the 2026-11-07 DB rename program (#6619), while repointing runtime-db-export and quota-recorder across machines.
-
The LIVE copies are the sandbox ones. The paused runtime-db-export cron is a sandbox agent job whose body drives ~/tools/runtime-db-export (sandbox). The quota recorder's working home is also the sandbox (its FLEET path ~/tools/bin/fleet resolves there via symlink). Both were repointed to fleet-mirror / fleet-platform on 2026-11-07 and the sandbox hatch commits pushed; the laptop ~/hatch clone fast-forwarded cleanly to the same code.
-
Latent break: laptop ~/hatch/tools/quota-recorder.py resolves its secrets via FLEET = ~/tools/bin/fleet, which does NOT exist on the laptop (the laptop binary is ~/.local/bin/fleet). A manual laptop run on 2026-11-07 failed with turso-error FileNotFoundError: '/home/user/tools/bin/fleet' and left its record pending. Harmless today (the laptop copy is not the live home; the JSONL is backup-of-record and the record stays pending), but anything that ever runs the recorder on the laptop silently stops landing quota records. Its shipped systemd unit is also stale (HOME=/home/hatch, 'Supabase landing' description, timer not installed).
-
~/git/hatch-dispatch-clone/tools/runtime-db-export on the laptop is a stale PARTIAL copy: no turso_client.py at all (pre-Turso layout), docstrings still naming muse-mirror. It cannot run standalone, but its presence invites exactly the 'which copy is live?' confusion the rename program had to burn time settling.
Fix direction: decide per tool which clone is canonical, delete or clearly mark the stale copies (dispatch-clone is a test fixture — if it must stay, pin it and label it), and make quota-recorder resolve the fleet binary by PATH lookup with a ~/tools/bin/fleet fallback instead of one hardcoded absolute path.
Acceptance criteria
- Each named tool (quota-recorder, runtime-db-export) has one explicitly named canonical clone, recorded in the issue resolution, and stale copies are deleted or clearly marked as non-live fixtures.
- quota-recorder resolves the fleet binary by PATH lookup with a ~/tools/bin/fleet fallback (no single hardcoded absolute path), and a manual laptop run lands a quota record without FileNotFoundError.
- The stale partial copy in ~/git/hatch-dispatch-clone is removed, or pinned and labelled as a test fixture that cannot be mistaken for the live copy.
- The shipped quota-recorder systemd unit is corrected (HOME, description) or removed if the timer is not installed.
Activity
developing-today commentedon Nov 7, 2026
CODE-2103
dezren39 commentedon Nov 7, 2026
Intake review
Related: #6619 (DB rename program, where this was found). No merged PR found in this pass that resolves the drift — the repointing in #6619 fixed DB names, not the laptop path/canonical-copy problem described here.
Notes: the failure is latent (the laptop copy is not the live home today), so normal priority. The concrete break is the hardcoded ~/tools/bin/fleet path in the laptop quota-recorder copy; the rest is canonical-copy hygiene.
Risks: deleting the dispatch-clone copy could break a test fixture that depends on its shape — verify before removal, otherwise pin + label it.
Suggested approach: decide canonical clone per tool first, fix the fleet-binary resolution in quota-recorder, then remove/mark stale copies and fix or drop the stale systemd unit.
🤖 Agent note · handed off
dezren39 commentedon Nov 8, 2026
🤖 Agent note · handed off
developing-today commentedon Nov 8, 2026
🤖 Agent note · handed off