Break down a Claude Code session's costs by assistant message to find expensive responses. Compare input, output, cache writes, cache reads, and models, with message cost percentiles providing context for investigating unusually costly activity.
Per-message cost breakdown within a single session. The drill-down companion to cost-anomaly — when an outlier session is flagged, this surfaces the specific expensive messages so operators can see whether the cost came from output tokens, cache writes, or model escalations.
Review the instructions the agent will receive before you install.
When cost-anomaly flags a session as a >3.5σ outlier, the next question is "which MESSAGES were expensive?". cost-session answers that.
| Question | Skill |
|---|---|
| "Which sessions cost the most?" | cost-conversation |
| "Which sessions are outliers?" | cost-anomaly |
| "Which messages in THIS session were expensive?" | cost-session ← this |
Implementation: scripts/session.mjs.
--session-id <id> (scans ~/.claude/projects/*/)
or --latest (default; picks most-recently-modified jsonl).usage blocks._prices.mjs).cost_usd, surface top-N (default 20).Example real session, top message:
| # | Model | In | Out | Cache W | Cache R | Cost |
| 1 | opus-4-7 | 6 | 569 | 881898 | 0 | $16.58 |
Without the Cache W column it looks like "569 output tokens cost $16" — that's wrong by 380×. The actual cost is ephemeral 1h cache write at opus pricing: 881,898 tokens × $18.75/1M = $16.54.
Operators reading the table see immediately: "the model wrote 881K tokens to ephemeral cache". From there the question becomes "why did we cache 881K tokens of context for a 6-input request?" — that's a real engineering signal.
# Step 1: find outliers across all sessions
cost anomaly --alert-on-outliers 1 || cost anomaly # see which session-ids
# Step 2: drill into the flagged session
cost session --session-id <flagged-id> --top 10
# Step 3: open that jsonl at the timestamp the top message reports,
# inspect the prompt + tool calls
Top of output:
| p50 (median) message | $0.85 |
| p90 message | $1.45 |
| p99 message | $1.74 |
Lets operators ask "is this top message a 2× outlier or a 380× one?" without having to compute it themselves. The "top is >2× p99" footer fires when the answer is "yes, this is an in-session outlier worth investigating".
Useful for drilling into a specific time range within a long session:
cost session --since 2026-06-16T13:00:00Z --top 5
Only messages with timestamp >= --since are considered.
--session-id not found in any project's jsonls → exit 2 with error.--top must be a positive integer → exit 2.No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Compare models available through the HuggingFace router with chat-ui configuration, research missing models, and write short English descriptions of their architecture and capabilities. Add entries to both production and development configuration files and commit the changes.
Create an Architecture Decision Record, a document for recording important design choices, with the next sequential number. It generates a Markdown template, registers the record in AgentDB, and links related decisions found through search.
Build a searchable index and relationship graph from ADRs, records of architectural decisions. Import two supported document formats, refresh existing metadata without duplicate relationships, and review summaries of statuses, missing references, and inconsistencies.
Rebuild the index of ADRs, records of architectural decisions, after files or relationship lines are deleted. It clears stored records and relationships, indexes the files still present, and checks that the final record count matches.
Review code changes against accepted Architecture Decision Records, documents that capture agreed design choices. Report violations, outdated references, compliant changes, and areas without ADR coverage, then suggest code fixes or a new decision record where appropriate.
Check saved architecture decision records (ADRs) for broken references, circular replacement relationships, and inconsistent statuses. Use the results to verify imported records or stop a CI check when selected integrity rules fail.