ac-status

name	ac-status
description	Show auto-context pipeline status including observation counts, candidates, conventions, anti-patterns, reward signals, and lifecycle statistics.
disable-model-invocation	true

Auto-Context: Pipeline Status

Display the full auto-context pipeline status dashboard. This is a read-only skill -- never modify any data files.

Important rules:

• Use jq for all JSON reading and counting.
• Handle missing fields with jq // alternative operator.
• If any jq command fails (corrupted JSON), show "Error reading {file}" instead of crashing.
• Never write to or modify any data files.

---

Step 1: Read Pipeline Data

Read the following files from .auto-context/ in the project root:

• conventions.json -- convention entries with stage field
• candidates.json -- candidate entries with stage and observations
• anti-patterns.json -- anti-pattern entries with source and stage fields
• lifecycle.json -- session counter and metadata
• config.json -- token budget configuration
• rewards.json -- per-session reward signal history

If the .auto-context/ directory does not exist, tell the user:

Auto-context is not initialized. Run /ac-init first.

And stop.

Handle missing individual files gracefully (treat as empty arrays or default objects).

---

Step 2: Gather Counts

Use these jq commands to gather counts:

# Session info
jq -r '.session_count // 0' .auto-context/lifecycle.json
jq -r '.token_budget // 1000' .auto-context/config.json

# Convention counts by stage
jq '[.[] | select(.stage == "active")] | length' .auto-context/conventions.json
jq '[.[] | select(.stage == "decayed")] | length' .auto-context/conventions.json

# Candidate counts by stage
jq '[.[] | select((.stage // "observation") == "observation")] | length' .auto-context/candidates.json
jq '[.[] | select(.stage == "review_pending")] | length' .auto-context/candidates.json
jq 'length' .auto-context/candidates.json

# Anti-pattern breakdown by source (treat missing .source as "explicit" for Phase 4 entries)
jq '[.[] | select((.source // "explicit") == "explicit")] | length' .auto-context/anti-patterns.json
jq '[.[] | select(.source == "correction")] | length' .auto-context/anti-patterns.json
jq '[.[] | select(.source == "error")] | length' .auto-context/anti-patterns.json
jq '[.[] | select(.stage == "active")] | length' .auto-context/anti-patterns.json
jq '[.[] | select(.stage == "observation")] | length' .auto-context/anti-patterns.json

# Reward signal data
jq 'length' .auto-context/rewards.json
jq 'if length > 0 then ([.[].reward_score] | add / length * 100 | floor / 100) else null end' .auto-context/rewards.json
jq 'if length >= 5 then [-5:] | [.[].reward_score | . * 100 | floor / 100] else [.[].reward_score | . * 100 | floor / 100] end' .auto-context/rewards.json
jq 'if length > 0 then {implicit_pos: ([.[].implicit_positive] | add // 0), implicit_neg: ([.[].implicit_negative] | add // 0), explicit_pos: ([.[].explicit_positive] | add // 0), explicit_neg: ([.[].explicit_negative] | add // 0)} else null end' .auto-context/rewards.json

# Observation stats
jq '[.[].observations // 0] | add // 0' .auto-context/candidates.json
jq '[.[].session_id] | unique | length' .auto-context/candidates.json

# Recent changelog (last 5 entries)
tail -5 .auto-context/changelog.jsonl

---

Step 3: Format Status Dashboard

Present the dashboard using this format:

## Auto-Context Pipeline Status

### Session Info
- **Session count:** {session_count}
- **Token budget:** {token_budget} tokens

### Conventions ({active_count} active)
- Active (in CLAUDE.md): {active_count}
- Decayed: {decayed_count}
- Convention cap: {active_count}/50

### Candidates ({total_candidates} total)
- Observations (gathering data): {observation_count}
- Ready for review: {review_pending_count}

If review_pending_count > 0, add: "Run /ac-review to review pending candidates."

### Anti-Patterns ({active_ap_count} active, {observation_ap_count} pending)
- From explicit feedback: {explicit_ap_count}
- From correction detection: {correction_ap_count}
- From error tracking: {error_ap_count}
- Active (in CLAUDE.md): {active_ap_count}
- Observation (gathering data): {observation_ap_count}

Reward Signals

If rewards.json is missing or empty (length == 0):

### Reward Signals
- Reward tracking: no data yet (accumulates over sessions)

If rewards.json has data:

### Reward Signals ({sessions_tracked} sessions)
- Average reward score: {avg_score} (1.0 = perfect, 0.0 = all corrections)
- Implicit signals: {implicit_pos} positive / {implicit_neg} negative
- Explicit signals: {explicit_pos} positive / {explicit_neg} negative
- Recent trend (last 5): {score1}, {score2}, ...
  Trend: {trend}

For the trend determination:

1. Take the recent scores array (last 5 or fewer)
2. Compare the first and last values
3. If last - first > 0.1: "improving"
4. If first - last > 0.1: "declining"
5. Otherwise: "stable"

### Observation Stats
- Total observations across all candidates: {sum_observations}
- Unique sessions with extractions: {unique_session_count}

### Recent Activity

Read the last 5 lines of .auto-context/changelog.jsonl. Format each line as:

- [{timestamp}] {action}: {text} ({reason})

If no changelog exists or is empty: "No lifecycle activity recorded yet."