Agent Cost Report skill | PlugMCP

Agent Cost Report

thedotmack115kstars

Knowledge MemoryDatabasesDeveloper Tools

1---2name: Agent Cost Report3description: >-4---56# claude-mem-grok-bot/skills/agent-cost-report/SKILL.md

SKILL.mdAgent skill

Transport: File
Status: Listed

SKILL.md

Install

The skill file stays in the publisher's repository.

Paste the copied command in a terminal. It installs this SKILL.md into that agent.

shterminal

1npx skills add thedotmack/claude-mem --skill agent-cost-report -a cursor -y

Paste the copied command in a terminal. It installs this SKILL.md into that agent.

shterminal

1npx skills add thedotmack/claude-mem --skill agent-cost-report -a claude-code -y

Paste the copied command in a terminal. It installs this SKILL.md into that agent.

shterminal

1npx skills add thedotmack/claude-mem --skill agent-cost-report -a codex -y

Paste the copied command in a terminal. It installs this SKILL.md into that agent.

shterminal

1npx skills add thedotmack/claude-mem --skill agent-cost-report -a grok -y

Send the copied message in a Grok Bot chat. The Bot fetches this SKILL.md from GitHub and adds it to the skills all your Bots share.

txtGrok Bot chat

1Install this skill: https://github.com/thedotmack/claude-mem/tree/HEAD/claude-mem-grok-bot/skills/agent-cost-report

Download skill claude-mem-grok-bot/skills/agent-cost-report/SKILL.md on GitHub

claude-mem-grok-bot/skills/agent-cost-report/SKILL.mdMarkdown · UTF-8

Skill facts

Agent skill

File: claude-mem-grok-bot/skills/agent-cost-report/SKILL.md
Publisher: thedotmack
Hosting

FAQ

Questions

How do I install Agent Cost Report?

From the publisher

SKILL.md

name: Agent Cost Report
description: - Believable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json.
allowed-tools: - Bash - Read - Write - AskUserQuestion - mcp__plugin_claude-mem_mcp-search__search - mcp__plugin_claude-mem_mcp-search__timeline - mcp__plugin_claude-mem_mcp-search__get_observations

Agent Cost Report

Claude-Mem / Claude Code skill. Runtime is the scripts/ pipeline (transcripts → tokens → dollars → Timing-style report) plus a progressive Mem Search review pass that confirms the drafted labels. The Notion draft is SPEC history only — never the product, never the runtime, never the ship vehicle.

Resolve the absolute directory containing this SKILL.md; all helper paths are relative to that directory. ${CLAUDE_SKILL_DIR} is the shortcut: python3 "${CLAUDE_SKILL_DIR}/scripts/acr.py" …. Python 3.9+ standard library only, with IANA timezone data for America/Los_Angeles; no pip installs. tzdata is the only exception to the no-pip-installs rule, and only when acr.py exits saying that timezone data is unavailable (stock Windows Python ships none): show the user the lines it printed and ask (AskUserQuestion) before installing it. On a yes, run the line for your shell (the plain quoted line in Bash or cmd, the PowerShell: line in PowerShell), then rerun the failed command. Setting PYTHONTZPATH instead needs no install. The look lives in scripts/acr/render.py, never here.

Purpose

Turn Claude-Mem activity into a manager-readable cost and failure report.

Product idea: a reusable skill that searches Claude-Mem via Mem Search, reconstructs real units of work, assigns cost and failure categories, and renders a printable report.

Insight north star

The headline is dollars, to two decimals, labeled. The dollars come from Claude Code transcripts (exact per-reply token usage) priced at OpenRouter public list prices, so they are ESTIMATED. Measured provider spend appears only when a sanctioned source gives it. Directly under the dollars: what the mistakes cost, what shipped, and both on one time axis.

Questions the report must answer

What work was completed?
What did each outcome cost?
What was wasted through looping, hedging, wrong turns, rework, or poor routing?
Were any unauthorized actions attempted?
What should the manager change next?

Primary unit = cost per completed outcome (not cost per observation).

When to use

agent_cost_i (per reply, micro-dollars) =
      input × price.input + output × price.output
    + cache_write_5m × price.cache_write + cache_write_1h × price.cache_write_1h
    + cache_read × price.cache_read                    # cache_write_1h = listed rate, else 2 × input

agent_estimated_usd        = Σ agent_cost_i over every reply in the window (matched or not)   # ESTIMATED, the headline
extrapolated_unmeasured    = observer tokens of sessions with no transcript here × (measured $ per observer token)
                           # EXTRAPOLATED (low confidence); "all sessions measured" when nothing remains
observer_note_taker_est    = note-taker (observer) tokens, deduped per reply, × its input list price
                           # priced separately, never agent cost, never in the headline
mistakes_estimated_usd     = Σ agent_cost_i over the same-session union of wasted turns, each turn once   # low figure
cost_per_completed_outcome = Σ attributed $ for status ∈ {shipped, completed} / count(those work items)
waste_rate                 = Σ wasted_cost / Σ attributed $     recovery_share = Σ recovery_cost / Σ attributed $

Label	Meaning
Measured	Provider-reported spend from a sanctioned source (today: the OpenRouter per-key snapshot, shown with its UTC bucket label)
Estimated	Measured transcript tokens × OpenRouter list price per MTok (the formula above)
Extrapolated	Sessions with no transcript on this machine, from the observer-token ratio; always "low confidence"
Unavailable	No measured value — write `measured spend unavailable`, never `$0 spent` for unknown
risk_exposure	Severity / qualitative unless real cash is at stake (seats, refunds, SLA)

Field	Notes
`work_item_id`	Stable id (`WI-1` …)
`title`	Human work-item title (ALWAYS invent at review; drafts use the session's first prompt)
`scope` / `project` / `worktree`	Period, session or project; primary project; worktree when relevant
`session_ids` / `content_session_id`	Mem session id(s) and the transcript join key
`status`	`shipped` / `completed` / `in_progress` / `abandoned` / `blocked`
`category` / `work_category`	One of the six work categories
`failure_type` / `failure_signals`	Empty or failure types (Rework lives here)
`cost_measured`	Number or `unavailable`
`cost_estimated` / `cost_extrapolated` / `attributed_usd`	List-price estimate; extrapolation for sessions with no transcript; the one used
`wasted_cost` / `recovery_cost` / `productive_cost`	USD from the behavior union (0 if none)
`risk_exposure`	`none` / `low` / `medium` / `high` (non-dollar unless real cash)
`evidence_ids` / `summary_ids`	Observation and summary IDs
`agent_tokens`	`{input, output, cache_write_5m, cache_write_1h, cache_read}`
`observer_tokens`	Note-taker tokens, Details only
`model` / `model_prices_usd_per_mtok`	Dominant model and the prices used
`cost_basis`	`estimated_usage` / `measured_provider` / `extrapolated`
`label_source`	`keyword` (draft) / `llm` / `human`, with `reviewed_by` and `reviewed_at_pt`
`device`	`local`, `remote-N`, or an export label such as `mac`
`behavior_counts`	Pattern hits in this session
`recommended_action` / `confidence` / `date_pt` / `notes`	One lever; high/medium/low; PT day; one plain sentence

OpenRouter key (needs Alex) — a regular inference OPENROUTER_API_KEY supplied as an environment variable through the house's secure secret flow; never a management or provisioning key, never read from a settings file, never printed or written to any output. The per-key endpoint gives UTC day/week/month snapshots, so the figure is shown with its bucket label and switches the hero to MEASURED only when the bucket fully covers the report window. A range-by-day source (/api/v1/activity) exists but needs a management key, so it is not implemented (G5).
Mac transcripts (needs Alex) — sessions from Alex's Mac have no transcript on the box, so their cost is EXTRAPOLATED (low confidence) until a device export is merged. Either Alex runs, on the Mac, from a plain checkout of this skill (system python3 3.9+): python3 scripts/acr.py collect --export-device mac --start YYYY-MM-DD --end YYYY-MM-DD --out ~/acr-export and shares ~/acr-export/device-usage-mac.json (ids, timestamps, token counts, model names; no prompt or observation text, no settings, no paths); or, only after Alex's explicit go for that specific run (window, machine, destination named), an orchestrating agent runs the same command on the registered Mac through the house's registered-machine tooling and copies only device-usage-mac.json to the box. A go for one run is not a go for the next (G6 covered Sep 18–26 only). Merge with --device-usage; joined sessions flip to estimated_usage with device: mac.
Grok Bot (ALWAYS unavailable, no seat count) — Grok Bot is Cursor's cloud agent. Cursor shows its weekly usage only on the plan screen; there is no API or export, and the house has no xAI key for it. The report says exactly "Grok Bot usage: unavailable" (G4): no dollars, no guessed per-seat price, no seat count. A figure Alex supplies by hand would be entered as measured_manual with entered_by: Alex and the date, never inferred; that would be a new decision.
Wins from GitHub (read-only) — besides box transcripts and ship observations, rollup lists merged PRs with gh pr list --state merged for the repos in --wins-repo (default thedotmack/claude-mem), filtered by merge time inside the PT window and deduplicated by repo and PR number. When gh is missing, unauthenticated or rate-limited, that source reads "unavailable" in Details and the report still completes; it is never a zero. --no-gh turns every gh call off.
Win cost (unmeasured until R4) — sessions are not linked to PRs today, so every win reads "unmeasured (session not linked to PR)". Once commits carry a Claude-Session: <content_session_id> trailer (G13: trailer only, bare id), PR and publish wins get ESTIMATED · session-linked costs. There is no fallback estimate (G11).
Merge to main, npm publish, /version-bump — each needs Alex's separate go. Pushing the work branch and opening the PR are routine.

Agent Cost Report

Install

Skill facts

Questions

How do I install Agent Cost Report?

Where does Agent Cost Report live?

Which agents can use Agent Cost Report?

SKILL.md

Agent Cost Report

Purpose

Insight north star

Questions the report must answer

When to use

Default scope (ALWAYS)

Progressive Mem Search (ALWAYS)

Work categories (ALWAYS)

Failure / waste types (ALWAYS)

Cost model

Behavior metrics (heuristic until reviewed)

Money labeling (ALWAYS)

Line-item schema

Deliverables (ALWAYS)

Manager-facing HTML sections (required order)

Truthfulness ALWAYS rules

Recipe (orchestrator)

Gaps and gates

Keeping copies in sync

Related

Creed