v0.11.3: judgment-layer hardening, single routing authority

[KodingDev]

v0.11.3

Output style

• Pillar 1 extends to the project's own artifacts: read the existing code, spec, and canonical data before writing competing logic; cost/performance/architecture decisions cite real measurements (settled invoices, logs, profiles) pulled first.
• Voice adds a calibrated-reporting register: once verification is exhausted, remaining conclusions are presented as bets — observed facts separated from inferences, confidence and blind spots named, each bet carrying what would prove it wrong.
• A new operational-discipline section prefers a reusable script over repeated one-off tool calls, and reads state at its source instead of polling.

Skills

• auto: the horizon for stepped-away work is honest readiness, not elapsed time — remaining verifiable work means the turn is not over. Blockers are earned (try another angle, dig deeper, dispatch a subagent first); a genuine fork is a decision only the human can make (taste, money, irreversibility).
• debug: evidence comes from instrumenting the real code path — logging added to the actual executable run with real data, never probe scripts or simulations. A fix must explain why the bug existed, or it is a workaround.
• review: the craft lens flags symptom patches — a guard, carve-out, or special case that narrows a visible failure while the cause remains, even when the diff improves behavior.
• execute: verification exercises the real flow with real data where a runtime surface exists.
• brainstorm: re-reads the user's provided material (request, plan, handoff, spec) in full before asking any clarifying question.
• delegate: model selection names the expensive failure — a wide fan-out at flagship pricing doing mechanical work a small model does identically — and the orchestrator reads the final synthesis itself: the conclusion is not delegable.
• respond: user feedback is trusted on intent and priorities but evaluated on technical merits like anyone's, per the Challenge Protocol.

Routing

• The session orientation is the single routing authority. The meridian router skill is removed: its content was a near-verbatim copy of the always-injected orientation, so the "invoke for the full reference" pointer never fired, and the two copies had drifted. The one line it carried that the orientation lacked (visual regressions are bugs even when they appeared one step ago in the same session) moves into the orientation.
• The ## Integration footers on every skill and research's dual-gate rationale section are removed — routing is owned by the orientation, skill bodies already name their invocations inline, and the rationale was a maintainer note on the prompt surface.

Release

• All three per-host manifests (.claude-plugin, .cursor-plugin, .plugin) move to 0.11.3; the CHANGELOG records the changes.