---
name: cost-report
description: Weekly API cost report — reads token usage CSV, computes dollar costs per skill and model, reports trends
var: ""
tags: [meta]
version: "1.0.0"
---
> **${var}** — Number of days to cover (default: 7). Pass "30" for a monthly view.

Today is ${today}. Generate a cost report from Aeon's token usage data.

## Model Pricing (per million tokens)

Use these rates to calculate costs. First check `aeon.yml` for the `gateway.provider` value.

### Direct Anthropic (gateway.provider: direct)

| Model | Input | Output | Cache Read | Cache Write |
|-------|-------|--------|------------|-------------|
| claude-opus-4-7 | $15.00 | $75.00 | $1.50 | $3.75 |
| claude-sonnet-4-6 | $3.00 | $15.00 | $0.30 | $3.75 |
| claude-haiku-4-5-20251001 | $0.80 | $4.00 | $0.08 | $3.75 |

### Bankr Gateway (gateway.provider: bankr)

| Model | Input | Output |
|-------|-------|--------|
| claude-opus-4-7 | $5.00 | $25.00 |
| claude-sonnet-4-6 | $3.00 | $15.00 |
| claude-haiku-4-5-20251001 | $0.80 | $4.00 |
| gemini-3-pro | $1.25 | $10.00 |
| gemini-3-flash | $0.15 | $0.60 |
| gpt-5.2 | $2.50 | $10.00 |
| kimi-k2.5 | $1.00 | $4.00 |
| qwen3-coder | $0.50 | $2.00 |

Note: Bankr does not break out cache read/write pricing separately. Treat cache columns as zero cost for Bankr rows.

For any model not in these tables, default to Opus pricing (conservative estimate).

## Steps

1. **Determine the report window.**
   - Default: 7 days. If `${var}` is set to a number (e.g. "30"), use that many days.
   - Compute `CUTOFF_DATE` = today minus N days.

2. **Read token usage data.**
   - File: `memory/token-usage.csv`
   - Columns: `date,skill,model,input_tokens,output_tokens,cache_read,cache_creation`
   - If the file does not exist: log "COST_REPORT_SKIP: no token-usage.csv yet" and stop — do NOT send any notification.
   - Filter rows where `date >= CUTOFF_DATE`.
   - If zero rows after filtering: log "COST_REPORT_SKIP: no runs in last N days" and stop.

3. **Compute costs for each row.**

   For each row, look up the model's rates and calculate:
   ```
   input_cost  = input_tokens  / 1,000,000 × rate_input
   output_cost = output_tokens / 1,000,000 × rate_output
   cache_read_cost    = cache_read     / 1,000,000 × rate_cache_read
   cache_write_cost   = cache_creation / 1,000,000 × rate_cache_write
   row_cost = input_cost + output_cost + cache_read_cost + cache_write_cost
   ```

4. **Aggregate into report sections.**

   a. **Total cost** for the window.
   b. **Per-skill cost table** — sum cost per skill, sort descending, show top 10:
      - Columns: Skill | Runs | Total Tokens | Cost | Avg Cost/Run
   c. **Per-model breakdown** — total cost and total tokens per model.
   d. **Token efficiency** — for each skill in top 10, compute output_tokens / total_tokens ratio (higher = more generative, less reading).
   e. **Week-over-week trend** — if window is 7 days and the CSV has ≥14 days of data:
      - Compare this week's total cost vs the prior 7-day period.
      - Show: "Cost up/down X% vs prior week"
   f. **Most/least efficient skills** (cost per 1K output tokens).

5. **Write the full report** to `articles/cost-report-${today}.md`:

   ```markdown
   # Aeon Cost Report — ${today}
   *Period: last N days*

   ## Summary
   - **Total API cost:** $X.XX
   - **Total tokens:** X,XXX,XXX (input: X | output: X | cache_read: X | cache_write: X)
   - **Skill runs tracked:** N
   - **Week-over-week:** ↑/↓ X% vs prior week  (omit if <14 days of data)

   ## Cost by Skill (Top 10)
   | Skill | Runs | Tokens | Cost | Avg/Run |
   |-------|------|--------|------|---------|
   | ...   | ...  | ...    | $... | $...    |

   ## Cost by Model
   | Model | Runs | Tokens | Cost |
   |-------|------|--------|------|
   | ...   | ...  | ...    | $... |

   ## Efficiency
   | Skill | Cost / 1K Output Tokens |
   |-------|------------------------|
   | ...   | $...                   |

   *Generated by Aeon cost-report skill*
   ```

6. **Send notification** via `./notify`:

   ```
   *Cost Report — ${today} (last N days)*

   Total: $X.XX across N skill runs

   Top 3 by cost:
   1. skill-a — $X.XX (N runs, avg $X.XX/run)
   2. skill-b — $X.XX
   3. skill-c — $X.XX

   By model:
   - Opus: $X.XX | Sonnet: $X.XX | Haiku: $X.XX

   Trend: ↑/↓ X% vs prior week  [omit if no prior week data]

   Full report: articles/cost-report-${today}.md
   ```

7. **Log** to `memory/logs/${today}.md`:
   ```
   ## Cost Report
   - Period: last N days
   - Total cost: $X.XX
   - Runs tracked: N
   - Most expensive skill: skill-name ($X.XX)
   - Cheapest skill: skill-name ($X.XX)
   - Week-over-week: +/-X%
   - Article: articles/cost-report-${today}.md
   - Notification sent via ./notify
   ```