efficiency

name	efficiency
description	Chief Efficiency Officer operator commands. Review proposals queued by the Stop-hook reflection layer (F019), read today's efficiency report, query operator velocity baseline, mute threshold-push interrupts. All observations are evidence-based — every proposal cites data points and operator baseline; no hallucinated narrative.

/efficiency — Operator commands for the Chief Efficiency Officer

The Chief Efficiency Officer (F019) is a Stop-hook reflection layer that observes operator efficiency, computes active-engagement against baseline, and writes evidence-cited proposals to a queue + daily report. This skill is the operator-facing command surface for reviewing those outputs.

The CEO does NOT block work. It observes, reflects, and proposes. The operator decides what to do with each observation.

Usage

/efficiency review    -- walk queued proposals one at a time, Pattern A
/efficiency today     -- print today's rolling efficiency report
/efficiency baseline  -- show operator velocity baseline (sdlc_timing.py)
/efficiency mute --until <iso8601> --reason "<text>"  -- silence threshold-push for a window

Workflow

/efficiency review

Reads .etc_sdlc/efficiency/proposals/*.md (oldest first). For each proposal, displays the data-cited observation to the operator and asks via Pattern A (AskUserQuestion):

AskUserQuestion(
  questions: [{
    question: "Proposal: <proposal title>. <data points summary>. What to do?",
    header: "Proposal",
    multiSelect: false,
    options: [
      {
        label: "Accept (move to accepted/, plan a response)",
        description: "The observation is correct and worth acting on. Moves the proposal to accepted/ for future reference. Operator decides the response separately."
      },
      {
        label: "Dismiss (false positive)",
        description: "The observation is wrong, the data is misleading, or there's a legitimate reason for the apparent anomaly. Moves the proposal to dismissed/ — kept for false-positive-rate tracking but not surfaced again."
      },
      {
        label: "Mark for follow-up feature",
        description: "The observation is correct AND warrants a harness feature. Creates a task in the task list referencing the proposal and moves the proposal to followups/."
      },
      {
        label: "Skip — review later",
        description: "Leave in the queue; come back to it."
      }
    ]
  }]
)

After processing each proposal, move to the next one. When the queue is empty, render the summary:

Reviewed N proposals. Accepted: A. Dismissed: D. Marked for follow-up: F. Skipped: S.

/efficiency today

Reads .etc_sdlc/efficiency/daily/<current-date>.md and prints it to stdout. If today's file doesn't exist yet, print: "No efficiency observations recorded today. Stop hook fires on every turn end."

/efficiency baseline

Invokes python3 ~/.claude/scripts/sdlc_timing.py --baseline and prints the output. Convenience wrapper.

/efficiency mute --until <iso8601> --reason "<text>"

Writes .etc_sdlc/efficiency/mute.yaml:

until: 2026-05-15T18:00:00Z
reason: "Long-running debug session on regression X — expecting > 3h active"
muted_at: 2026-05-15T10:00:00Z

Reason MUST be non-empty (re-ask via Pattern B if empty). The mute file is read by the CEO Stop hook on every fire — when present and until is in the future, threshold-push is suppressed (daily report still updates).

Constraints

• The CEO is observation-only. This skill does NOT modify hook behavior, re-tune thresholds, or alter the data layer. Configuration changes happen via env vars (CEO_IDLE_THRESHOLD_MINUTES, CEO_SIGMA_PUSH, etc.) on the operator's shell, not via this skill.
• Evidence-based discipline applies to operator review too. When dismissing a proposal as "false positive," the operator should glance at the cited data points — many "false positives" are actually correct observations the operator hadn't internalized.
• Dismissed proposals are NOT deleted. They're moved to dismissed/<date>.md for false-positive-rate tracking. If dismiss rate exceeds 50% of proposals, the CEO thresholds are wrong and need tuning (operator responsibility for v1; auto-tuning is v2).

See also

• F019 spec: spec/chief-efficiency-officer.md
• hooks/chief-efficiency-officer.sh — the Stop hook
• hooks/sandbox-bypass-tracker.sh — PreToolUse hook for bypass counting
• scripts/sdlc_timing.py — operator velocity baseline data source
• Anthropic How Claude Code Works in Large Codebases — the Stop-hook reflection pattern this skill consumes