docs(context): model policy settled (Opus 4.8/medium) + task-descriptions-decay warning

Bake-off is closed — Opus 4.8/medium is the standing default, Sonnet arms
cancelled, per-task logging retired. Cost is no longer a model-selection
input: billing is hard-capped, so Opus's real cost is Pro daily-limit hits,
an accepted trade. Max is the fix, not a downgrade.

Also adds the warning that bit twice in one session: task descriptions and
handoffs decay. #149's "delete the stale playbook" hazard was already done
the night before, and session_handoff.md had been asserting a FALSE
passwordless-sudo alarm plus a DEAD root token as live into every session
for 3 days. Check the artifact before doing the work.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Backtalk6858
2026-07-15 13:24:12 -05:00
parent c3b4295e63
commit dcc4cc4ef2
+12 -1
View File
@@ -7,13 +7,24 @@ project_name: agent-builder
> current state from it — it is 568 lines of June sprint-day facts with the oldest on top.
> Fix pending: prune it (task #192). See [[feedback-superseding-records-need-a-pointer]].
## CURRENT STATE (updated 2026-07-14)
## CURRENT STATE (updated 2026-07-15)
**What this project actually is now:** the home for **Agent-Sudo** + **Constrained Autonomy (CA)**
the system that removes the user's rubber-stamp permission tax by replacing approval gates with
enforced deterministic constraints. The original N8N Agent Builder / N8N Builder Agent framing below
is **historical**; those are not the active work.
**Model/effort: Opus 4.8 / MEDIUM** — settled 2026-07-15. Bake-off **closed**, Sonnet arms
**cancelled**, per-task bake-off logging **retired**. Billing is hard-capped (**>$20/mo
impossible**), so cost is not a model-selection input; Opus's real cost is more frequent Pro
daily-limit hits, an explicitly accepted trade. **Do not propose Sonnet; do not re-raise cost.**
**⚠️ TASK DESCRIPTIONS AND HANDOFFS HERE DECAY — CHECK THE ARTIFACT BEFORE DOING THE WORK.**
On 2026-07-15 this bit twice in one session: #149's "HAZARD: delete playbook_agent_sudo_phases" was
already done the night before, and `session_handoff.md` was asserting a **false** security alarm plus
a **dead** root token as live, into every session, for 3 days. Prune = **#192**; recurring guard =
`automation_ideas` **id=92**. See [[feedback-superseding-records-need-a-pointer]].
**Live right now:**
- **Agent-Sudo is deployed and ENFORCING on both hosts** (#145 closed 2026-07-14).
`SUDO_MD_VERIFY_ENFORCE=true`, signed digest `38eca778abb0..`, daemon `NRestarts=0`.