loops: script the deterministic lane mechanics — machine-behavior block, #502 telemetry upsert, wire script-the-deterministic-work into prompt ritual

[kyle-sexton]

Category: deterministic-work extraction · Source: control-tower gap-hunt 2026-07-19 (skill-catalog audit)

Problem — lane prompts make the model re-do pure mechanics every cycle

Standing operator principle: what needs no reasoning gets a script. Three concrete violations in the current loop prompts:

1. MACHINE-BEHAVIOR telemetry block — every lane hand-assembles worktree root/pattern/count, clone path, gh identity, plugin versions, deviations each session. Pure environment inspection; machine-health / claude-config:audit already inspect this surface. Fix: script emits the block; prompt references the script.
2. ops: loop-lane telemetry — running per-lane status (babysit / triage / work) #502 telemetry-comment upsert — "maintain ONE comment, EDIT in place, never post a second" needs a prompt warning against double-posting — the tell that marker-based edit-in-place belongs in a script. Folds into the WP2 telemetry contract that supersedes ops: loop-lane telemetry — running per-lane status (babysit / triage / work) #502; script is the interim.
3. Prompt-versioning ritual never runs re-anchor:script-the-deterministic-work over loop-prompts/ — a skill exists whose whole job is finding exactly these extractions; wire it into the prompt-authoring ritual (and feat(claude-ops or session-flow): loop-prompt authoring skill — generate and maintain standing loop-lane prompts #480's future skill) so this class self-detects instead of needing an audit to spot it.

Effect

Shrinks lane prompts (context-minimalism: long prompts = micromanagement, and Fable-class models need fewer steers), removes per-cycle token spend on mechanics, and makes the outputs deterministic/testable.

Relations

#502 (interim home of the telemetry block), #480 (prompt-authoring skill = ritual home for item 3), #477-#479 (absorption umbrellas — these extractions are the script-shaped subset), #496 (context economy).

[kyle-sexton]

This was generated by AI during triage. · Category: deterministic-work extraction · Priority: priority: high (pipeline throughput + the operator's "what needs no reasoning gets a script" principle; removes per-cycle token spend + makes outputs testable). Routed status: ready.

Plan

Three concrete deterministic extractions out of the loop prompts:

1. MACHINE-BEHAVIOR block → script. A script emits worktree root/pattern/count, clone path, gh identity, plugin versions, deviations (pure environment inspection — machine-health / claude-config:audit already inspect this surface); the prompt references the script instead of hand-assembling the block each session.
2. ops: loop-lane telemetry — running per-lane status (babysit / triage / work) #502 telemetry-comment upsert → script. Marker-based find-or-create + edit-in-place ("maintain ONE comment, EDIT in place, never double-post") is exactly the deterministic pattern that belongs in a script (the prompt currently needs an anti-double-post warning — the tell). Interim home ops: loop-lane telemetry — running per-lane status (babysit / triage / work) #502; folds into the WP2 telemetry contract that supersedes it.
3. Wire re-anchor:script-the-deterministic-work into the prompt-authoring ritual — a skill exists whose whole job is finding these extractions; run it over loop-prompts/ (and feat(claude-ops or session-flow): loop-prompt authoring skill — generate and maintain standing loop-lane prompts #480's future authoring skill) so this class self-detects instead of needing an audit.

Acceptance

• The machine-behavior block and the ops: loop-lane telemetry — running per-lane status (babysit / triage / work) #502 upsert are produced by scripts the prompt invokes; no per-cycle hand-assembly.
• re-anchor:script-the-deterministic-work is a step in the prompt-authoring ritual / feat(claude-ops or session-flow): loop-prompt authoring skill — generate and maintain standing loop-lane prompts #480.
• Lane prompts shrink (context-minimalism — long prompts micromanage; Fable-class models need fewer steers).

Live-lane confirmation: the triage lane is hand-assembling exactly these (machine block + telemetry upsert) every cycle right now — this extraction removes that. Relations: #502 (interim telemetry home), #480 (ritual home for item 3), #477-#479 (absorption — these are the script-shaped subset), #496 (context economy).

Decision defaulted: extract the machine-block + telemetry-upsert into scripts and wire script-the-deterministic-work into the prompt-authoring ritual — veto before merge