Repository navigation
## v1.2.0 (2026-11-11)
v1.2.0 (2026-11-11)
Thirtieth agent: schedule-analyst, the school-and-schedule lane,
promoted from the watch list by the suite-completion gap survey. The
survey's baseline named weekly school work (td as the task system,
gcalcli the calendar, the Sunday D2L sweep) as the one recurring domain
with no seat; the same survey found the write axis saturated, and its
other verdicts are recorded in the roadmap ledger and watch list.
- td 5.10.0 is the verified backbone (every rung probed live
2026-11-11): today/upcoming/inbox, stats (live but absent from the
top-level help listing), task list's filter surface with the
results/nextCursor JSON pagination, the project/section/label
structure rungs, and completed list. Two quirks carried in the
charter: the JSON priority field is the API scale, inverted against
the UI (4 is p1), and today and upcoming open with the same overdue
block. - gcalcli 4.5.1 is chartered in its verified state:
listanswers from
local cache, but event queries currently fail with
RefreshError: invalid_grant(the stored OAuth session cannot
refresh; verified live). Until Brandon re-auths, the digest runs
td-only and carries a calendar-unavailable line with the re-auth
direction, the ledger-analyst precedent (chartered ready, the dormant
dependency stated plainly). Re-authentication is never the agent's
act. - Read-only by charter: td's and gcalcli's read verbs only,
recommendations never executed, recurrences kept visible in any
suggestion because they are load-bearing. The modes-before-agents
call is recorded: not data-analyst's named-endpoint lane (local
datasets, not CLI state reads) and not syshealth-auditor's (machine
state shares no corpus or allowlist with task and calendar APIs). - Model pin: GLM-5.3-Flash, the rubric-driven evidence-gathering tier
(syshealth-auditor precedent); the dispatcher can raise it if
digests run thin. - Packaging: roster check extends to thirty; README, AGENTS.md (whose
agent list carried a stale twenty-eight count over twenty-nine
names; fixed here), both manifests, and the roadmap synced. The
roadmap records the survey's verdicts: catalog reconciler declined
(no git baseline under the write target for the fence model's
re-verify leg, verified live), dependency apply mode and kindle-sync
deferred with entry conditions, emulation and ML seats declined as
job-title candidates, and ctf-recon and log-miner keep their unfired
conditions.