rtk-integration

harness-config shell layer

RTK (Rust Token Killer) owns the shell dedup token-efficacy layer. Repo policy lives in config/rtk-integration.json. Do not add @RTK.md to shared instruction surfaces — use wagents rtk instead.

Canonical config

config/rtk-integration.json declares per-harness RTK init commands, tiers, and notes. Edit policy here; run doctor/sync to verify projections.

uv run wagents rtk doctor --format json
uv run wagents rtk sync --dry-run --platforms claude-code,cursor,opencode,codex,gemini-cli,github-copilot
RTK_TELEMETRY_DISABLED=1 uv run wagents rtk sync --apply --platforms claude-code,cursor,opencode,codex,gemini-cli,github-copilot
uv run wagents rtk gain --graph
uv run wagents rtk gain --history
uv run wagents rtk review --format json

Stack sync integration

After hook projection, optionally chain RTK fleet init:

uv run python scripts/sync_agent_stack.py --apply --targets repo --with-rtk
# or
RTK_ENABLED=1 uv run python scripts/sync_agent_stack.py --apply --targets repo

No @RTK.md in shared corpus

Shared instruction files (instructions/global.md, AGENTS.md, generated mirrors) must not @-include RTK.md. Validation fails on stale includes; RTK posture is CLI-driven via wagents rtk doctor and wagents rtk sync.

Usage-review lane

wagents rtk review aggregates rtk gain --history and summary gain output into harness-master usage-review recommendations (instrument, tune-workflow, keep). Pair with /harness-config/token-efficacy/ for layer ownership and decision gates.

Related

token-efficacyEight-layer token posture hub and one-tool-per-layer rules.

HooksFleet hooks run before RTK shell projections when both apply.