feat(agentic): ranked reasoner search + ambient load metadata (beta-plan: Santosh's recommendations)

[AbirAbbas]

Implements the two remaining beta-plan recommendations ("Santosh's recommendations") for the sub-harness flow, where a coding agent (Claude Code, Codex, …) drives AgentField through the af agent JSON CLI and the agentfield-use skill.

1. Reasoner discovery at scale — ranked search

With hundreds of reasoners installed, the driving agent must search, not list-and-scan.

• GET /api/v1/agentic/reasoners?q=<free text>&agent=<optional id>&limit=<1..50> — ranks all registered reasoners with a dependency-free BM25F-lite index built per request from the same registration data the discovery surface serves (results never drift from discovery). Field boosts: reasoner id 3.0 > tags 2.0 > agent id 1.5 > metadata description 1.0; the tokenizer splits snake/kebab/camelCase (query "pr resolve" matches run_pr_resolver); ties break deterministically by invocation target.
• Each hit carries reasoner_id, agent_id, invocation_target, tags, score, agent_health — the brain can dispatch immediately with no second lookup.
• CLI: af agent search "<free text>" [--agent <id>] [--limit N].
• Empty q → structured missing_query error.

2. Ambient machine-load metadata — zero extra round-trips

The brain must know how loaded the machine is before starting more heavy agents, without asking.

• Every agentic response now carries meta.load = {running_agents, total_agents, active_executions, cpu_cores, recommended_max_concurrent}, stamped centrally in respondOK via a load provider registered at route setup.
• Provider caches for 2s so bursts don't hammer storage; on any error meta.load is silently omitted — it can never fail a response. recommended_max_concurrent = max(1, cores/2) (cores-based; memory-aware refinement noted in-code as follow-up).
• The af agent CLI already forwards server meta into its JSON output, so driving agents get this on every call for free.

Skill wiring

agentfield-use v0.2.0 → v0.3.0 (embedded mirror synced, sync-embedded-skills.sh --check passes):

• "Too many reasoners to scan? Search, don't dump" — use af agent search past ~20 reasoners; dispatch from invocation_target; prefer agent_health == "active".
• "Check the load before piling on" — read meta.load; if active_executions >= recommended_max_concurrent, finish in-flight work before launching more, and tell the user when throttling.

Testing

• 20 new unit tests (BM25 ranking/tokenizer, reasoners endpoint incl. filters/clamps/error paths, load-provider caching/fallback, envelope marshaling, CLI subcommand mapping) — internal/handlers/agentic and targeted internal/cli / internal/skillkit suites green.
• Runtime-verified end-to-end on a live control plane with 78 registered reasoners across 4 agents: af agent search "review pull request" ranks pr-af-go:review_dimension / pr-af-go:review on top; --agent swe-planner + "plan a development sprint" returns swe-planner:plan then run_sprint_planner; meta.load present on every response (16-core machine → recommendation 8); empty query returns the structured error.

Notes

• No new module dependencies; corpus rebuilt per request (small population) — no background indexing.
• Known pre-existing Windows-only test quirks in untouched files (internal/cli captureOutput pipe-buffer deadlock, skillkit /tmp/HOME assumptions) are unaffected; Linux CI is the arbiter.

🤖 Generated with Claude Code

[github-actions]

📊 Coverage gate

Thresholds from .coverage-gate.toml: per-surface ≥ 84%, aggregate ≥ 85%, max per-surface regression ≤ 1.0 pp, max aggregate regression ≤ 0.50 pp.

Surface	Current	Baseline	Δ
control-plane	86.80%	87.40%	↓ -0.60 pp	🟡
sdk-go	92.50%	92.00%	↑ +0.50 pp	🟢
sdk-python	93.82%	93.73%	↑ +0.09 pp	🟢
sdk-typescript	90.44%	90.42%	↑ +0.02 pp	🟢
web-ui	84.76%	84.79%	↓ -0.03 pp	🟡
aggregate	85.51%	85.75%	↓ -0.24 pp	🟡

✅ Gate passed

No surface regressed past the allowed threshold and the aggregate stayed above the floor.

[github-actions]

📐 Patch coverage gate

Threshold: 80% on lines this PR touches vs origin/main (from .coverage-gate.toml:thresholds.min_patch).

Surface	Touched lines	Patch coverage	Status
control-plane	337	83.00%	✅
sdk-go	0	—	➖ no changes
sdk-python	0	—	➖ no changes
sdk-typescript	0	—	➖ no changes
web-ui	0	—	➖ no changes

✅ Patch gate passed

Every surface whose lines were touched by this PR has patch coverage at or above the threshold.