| name | score |
| description | Track and analyze the Autonomy Score — the experiment's primary KPI. Measures what percentage of Mentilead operational decisions required no human input over a rolling period. Three modes: update decision counts, review the current score with chapter-appropriate context (5-second check), or analyze trends across months. Use when updating the autonomy log, checking the score, or analyzing decision patterns over time.
|
| disable-model-invocation | true |
GrowthOS Autonomy Tracker
Step 0: Detect & Route
Before starting any mode, determine the current state and route to the right step.
Read Current State
- Read
marketing/MEMORY.md and marketing/STATUS.md
- If
marketing/ does NOT exist: tell the user to run the init skill first and stop
- Read
references/experiment-thesis.md — required by the Experiment Layer rules in CLAUDE.md
- Check if
marketing/experiment/autonomy-log.md exists. If not, create it from templates/autonomy-log.md
- Read
experiment_chapter from STATUS.md frontmatter (default: "Chapter 1: Foundation")
- Parse the Monthly Summary table to get the current month's row and previous month's row
Route by Intent
Detect intent from the user's message and route:
- Mentions logging/adding/recording decisions → Step 1
- Mentions score/check/where do we stand → Step 2
- Mentions trend/history/over time/chart → Step 3
Present Options
If the user's intent is not clear from their message, present these options:
Autonomy Tracker — {experiment_chapter}:
Current month: {YYYY-MM}
Score: {score}% ({autonomous} autonomous / {total} decisions)
Trend: {direction} vs last month ({prev_score}%)
1. Update Log — record additional decisions
2. Score Review — current score with context
3. Trend Analysis — 6-month rolling view
Route to the appropriate step based on the user's choice or detected intent.
Step 1: Update Log
Batch-update decision counts for the current month. Works alongside the observation logger — adds to existing counters, never replaces.
Gather Counts
Read current month's row from marketing/experiment/autonomy-log.md Monthly Summary table.
If no row for the current month exists, show current zeroed state.
Ask each question in sequence:
- "Any additional autonomous decisions to log?" (number, or "none")
- "Any additional human escalations?" (number, or "none")
- "Any agent failures?" (number, or "none")
For each non-zero answer, optionally ask: "Which system/agent?" (for system breakdown tracking)
Update Monthly Summary Table
Update marketing/experiment/autonomy-log.md:
- Find or create the row for the current month (format:
YYYY-MM)
- Add to existing counters — do NOT replace (the observation logger may have already incremented)
- Recalculate Total:
autonomous + human_escalations
- Recalculate Score:
autonomous / total * 100 (rounded to nearest integer, format as {score}%)
- Agent Failures are tracked but NOT part of the score denominator (per thesis definition: score = autonomous / (autonomous + escalations))
Update System Breakdown
Update or create the System Breakdown section for the current month in marketing/experiment/autonomy-log.md.
The system breakdown section for each month follows this format:
### {YYYY-MM}
| System | Autonomous | Escalated | Failures |
|--------|-----------|-----------|----------|
| {system_name} | {n} | {n} | {n} |
If a section for the current month already exists, update counters for existing systems or add new rows. If no section exists, create one.
Optional Notes
Ask: "Any notes for this month?" (free text, appended to Notes column)
If the user declines, leave the Notes column as-is (preserve existing value or use —).
Confirm to User
Updated {YYYY-MM}:
Autonomous: {n} | Escalations: {n} | Failures: {n}
Score: {score}% ({direction} vs last month)
Session Log & State Update
- Append to
marketing/logs/{YYYY-MM-DD}.md:
- **Autonomy Tracker** — Updated {YYYY-MM} log: {autonomous} autonomous, {escalations} escalations, {failures} failures → {score}%
- Update
marketing/STATUS.md: set last_updated to today's date, preserve all other fields
Step 2: Score Review
Read-only display of the current Autonomy Score with chapter-appropriate context.
Read Data
- Read
marketing/experiment/autonomy-log.md (both Monthly Summary table and System Breakdown sections)
- Read
references/experiment-thesis.md for chapter expectations
- Parse current and previous month rows from the Monthly Summary table
Display Score
Autonomy Score — {Month Year}
Autonomous decisions: {N}
Human escalations: {N}
Agent failures: {N}
Score: {X}%
Trend: {up/down/stable} vs last month ({prev_score}%)
Chapter: {experiment_chapter}
Breakdown by system:
| System | Autonomous | Escalated | Failures |
|--------|-----------|-----------|----------|
| {sys} | {n} | {n} | {n} |
Chapter Assessment
Compare the score against expected trajectory from references/experiment-thesis.md:
- Chapter 1 (Foundation): Establishing baselines, score expected to be low or undefined
- Chapter 2 (Delegation): Score should start climbing as handoffs happen
- Chapter 3 (Autonomy): Score should be meaningfully high (target 60%+)
- Chapter 4 (Stress Testing): Score may dip as edge cases surface
- Chapter 5 (Verdict): Final measurement
Flag if score is significantly above or below expected range:
Chapter Assessment: {ON TRACK / AHEAD OF EXPECTED / BELOW EXPECTED}
{one-line explanation based on chapter context}
Chapter expectations are soft guidelines, not hard thresholds. The assessment flags deviation but does not alarm — the experiment is exploratory.
Session Log
Append to marketing/logs/{YYYY-MM-DD}.md:
- **Autonomy Tracker** — Score review: {score}% for {YYYY-MM}, {chapter_assessment}
This mode is read-only — no STATUS.md write needed.
Step 3: Trend Analysis
Multi-month view of autonomy score progression with escalation driver analysis.
Read Data
- Read full
marketing/experiment/autonomy-log.md
- Read
marketing/experiment/observations.md (for linking top observations to score changes)
6-Month Rolling Chart
Display an ASCII chart of the most recent 6 months (or all available months if fewer than 6):
Autonomy Score — 6-Month Trend:
Month | Score | Bar
----------|-------|------------------------------------------
{YYYY-MM} | {n}% | {'#' * (score / 2)}
{YYYY-MM} | {n}% | {'#' * (score / 2)}
...
Each # represents 2 percentage points. A score of 60% shows 30 # characters.
Top Escalation Drivers
Aggregate system breakdown data across all tracked months to identify which systems generate the most escalations:
Top Escalation Drivers:
1. {system}: {n} escalations ({pct}% of all escalations)
2. {system}: {n} escalations ({pct}%)
3. {system}: {n} escalations ({pct}%)
Key Observations
Surface the top 3 observations that most influenced the autonomy score. Look in marketing/experiment/observations.md for:
- Observations with reflections (highest value — they have the "why")
- Observations of type
autonomous-action, human-escalation, or agent-failure
- Among failures, prioritize highest severity
Key Observations:
- OBS-{NNN} ({date}): {one-line summary} [{type}]
- OBS-{NNN} ({date}): {one-line summary} [{type}]
- OBS-{NNN} ({date}): {one-line summary} [{type}]
Session Log
Append to marketing/logs/{YYYY-MM-DD}.md:
- **Autonomy Tracker** — Trend analysis: {n} months tracked, current score {score}%
This mode is read-only — no STATUS.md write needed.
Reference Table
| Mode | References Read | User Files Read | User Files Written |
|---|
| Step 0 | references/experiment-thesis.md | MEMORY.md, STATUS.md, experiment/autonomy-log.md | — |
| Step 1 | — | experiment/autonomy-log.md | experiment/autonomy-log.md, STATUS.md, logs/{today}.md |
| Step 2 | references/experiment-thesis.md | experiment/autonomy-log.md | logs/{today}.md |
| Step 3 | — | experiment/autonomy-log.md, experiment/observations.md | logs/{today}.md |