Review: should PAI adopt the Claude Code Advisor Tool (experimental)?

[virtualian]

Context

Claude Code ships an experimental Advisor Tool (per Claude Code changelog: "Advisor Tool (experimental): dialog now carries an 'experimental' label, learn-more link, and startup notification when enabled"). A recent release fixed a class of bugs where sessions got stuck with Advisor tool result content could not be processed errors on every prompt and /compact, but the feature remains tagged experimental.

Question for PAI: should we use it, ignore it, or actively avoid it?

What to review

1. Capability: what does the Advisor Tool actually do in current Claude Code? Read the learn-more doc it links to; the behaviour is likely to have moved since any initial announcement.
2. Overlap with existing PAI skills: does it duplicate Thinking (council mode), the Agents skill (custom-role advisors), or any Algorithm-phase guidance pattern? If yes, what's unique?
3. Fit with PAI patterns: the Advisor Tool is dialog-interactive; PAI's Algorithm is output-format-mandated (NATIVE/ALGORITHM/ITERATION/MINIMAL). Does enabling the Advisor Tool break the mandatory-output-format rule, or coexist?
4. Experimental risk: depending on an experimental feature means API/behaviour may change without notice and the feature may be withdrawn. Can PAI absorb that, or would it create brittle integrations?
5. Enablement stance: off by default / opt-in via settings.json / recommended-on / discouraged — pick one.

Decision criteria

• Adopt if it provides capability not achievable via existing PAI skills and experimental status does not jeopardise PAI reliability.
• Reject if it duplicates existing skills with no unique benefit, or experimental instability threatens session reliability.
• Defer (re-review in 60 days) if the feature is visibly in flux and a later read will be more productive.

Deliverable

• A decision note recorded somewhere durable in the PAI repo (location TBD — no existing ADR dir).
• Any resulting config change in settings.json (enable / disable flag) with rationale.
• If adopted: a one-paragraph note in the PAI skill/algorithm docs explaining when to reach for the Advisor Tool vs existing Thinking / Agents skills.

Out of scope

• Building PAI features that replicate Advisor Tool capability.
• A broader sweep of all Claude Code experimental features (track separately if wanted).