docs: specify token-efficient monitors

[kyleseaman]

Stack: #5171 → #5172 → #5173 → #5174 → #5175 → #5176 → #5177

Position: 1 of 7. Base: main. Next: #5172.

Defines the P0 contract for token-efficient, session-owned monitors and the delivery plan for the dependent PRs.

Verification: docs, brand, scrub, and stack boundary gates.

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of c3995d92828b3befeb3b89b3ebc3a2f26e3dcb6b and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] c3995d9

False positive or not applicable? A repository writer can comment:
/ai-review override gpt c3995d92828b3befeb3b89b3ebc3a2f26e3dcb6b: <one-sentence reason>

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed c3995d92828b3befeb3b89b3ebc3a2f26e3dcb6b — this comment is updated in place on each push.

Review details

This is a documentation-only PR: one new RFC file and one index row in the directory README. There is no executable code, no schema, no security boundary, and no behavioral contract in the diff. The new doc is indexed in README.md (docs-lint satisfied), CHANGELOG.md is untouched (the changelog rule is not implicated), and the front-matter/index dates are internally consistent. Nothing here supports a grounded (a)/(b)/(c) chain.

No findings.

[OPUS-REVIEWED] c3995d9

Verdict parsed from the review's SHA-scoped output markers for commit c3995d92828b3befeb3b89b3ebc3a2f26e3dcb6b.

False positive or not applicable? A repository writer can comment:
/ai-review override fable c3995d92828b3befeb3b89b3ebc3a2f26e3dcb6b: <one-sentence reason>

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of c3995d92828b3befeb3b89b3ebc3a2f26e3dcb6b — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

A real, evidenced cost problem (capped babysit loops, voluntary completion) solved at its root — server-owned probe/decide/wake — with compat, budgets, and rollout all specified.

The RFC's factual premises check out against the tree (the .babysit-key-* sidecar the scheduler never reads, max_cycles: 0 = unlimited, the cycle-cap terminal reasons in autonudge.py), the docs-only diff matches the stated scope, the index row lands in the same commit, and the alternatives section genuinely kills the nearby options (cron, workflow engine, cheaper model, arbitrary-command probes) rather than listing them. Versioned optional payload on NudgeLoop with fail-closed unknown versions and no eager migration is the right reversibility posture for PR 0 of a stack.

[DESIGN-REVIEWED] c3995d9

[kyleseaman]

Replaced by #5180 after renaming the branch to remove the prefix.