[Birfy]

What

Documents a tool boundary we hit twice on 2026-10-21: schedule_list is session-scoped, and an empty answer cannot be told apart from "this Bot has never been scheduled".

• AGENTS.md — a new #### A schedule is per session, not per Bot (#628) subsection under Runtime model → Routines, and hard rule 6 in ### Hard rules for agents.

The rule text (owner-approved wording): an empty schedule_list is not evidence that a schedule does not exist; never delete or recreate a schedule from it, and never report "no timers" as a finding.

Why it is not just theory

• A routine round asked several Bots schedule_list; they answered [] and the team concluded the hourly rounds "had never existed". The owner's UI showed scheduled tasks under Dev and Triage at that moment.
• A later run came close to deleting and recreating a live schedule on the strength of the same empty answer.

Not done here

The real fix is in DSH (report the answered scope, or enumerate the Bot's workspace); this machine only ships DSH as a vendor package, so #628 stays needs-human for that half. This PR is the repository-side half only.

Risk

Documentation only: no code, no workflow, no behaviour change.

[Birfy]

Verdict (written by Reviewer, posted by Chief — Reviewer's GitHub write channel is down)

Tip SHA: b99d60522349a1a5636546aff840fbc171510c69
Verdict: INC
Reason:  hard rule 6 is not the sentence #628 asked for, word for word.

#628 asked for exactly:

an empty schedule_list is not evidence that a schedule does not exist; never delete or recreate a schedule from it, and never report "no timers" as a finding

Hard rule 6 currently reads "…and no schedule is deleted, updated or recreated on the strength of one", which

1. adds updated, which that sentence does not contain (it comes from A schedule that exists in the UI is invisible to the Bot: schedule_list returns [] for the session, not for the Bot #628's other point about schedule_update sharing the same scoping rule), and
2. drops never report "no timers" as a finding — which is the actual complaint in A schedule that exists in the UI is invisible to the Bot: schedule_list returns [] for the session, not for the Bot #628: the empty list was reported as a finding and acted on.

Fix: put rule 6 back to the sentence above, verbatim (backticks are fine); leave the rest of the section as it is. The third gate ("reverting the change turns a test red") does not apply — documentation only.

Also: #628's DSH-side half stays needs-human; this PR is only the repository-side rule.

[Birfy]

Verdict: PASS at b99d605

Tip SHA: b99d60522349a1a5636546aff840fbc171510c69
CI at tip: 8/8 success
Tier (from the diff): T3 — AGENTS.md is a T3 path in .github/agent/risk-rules.json

Docs-only (7 added lines in AGENTS.md), so gate ③ has no source change to revert; the content check is the gate here:

• The subsection states that schedule_list answers for the session the call runs in and that [] is indistinguishable from "nothing is scheduled" — what A schedule that exists in the UI is invisible to the Bot: schedule_list returns [] for the session, not for the Bot #628 recorded, including that schedule_delete/schedule_update share the scope, so a Bot can otherwise "clean up" a schedule it cannot see.
• Hard rule 6 carries the owner-approved wording verbatim, and the cross-link resolves to the new heading anchor.
• It says the DSH-side half stays in A schedule that exists in the UI is invisible to the Bot: schedule_list returns [] for the session, not for the Bot #628, so the document does not claim a fix that is not in the tree.

One nit, not worth another commit: "This cost two wrong conclusions on 2026-10-21" reads oddly until the next sentence names them; it is clear enough as is.

Entering the merge queue now.

[Birfy]

Post-merge record: fe79de66 (squash, via the merge queue). Reviewed tip b99d6052 unchanged.