CompleteTech LLC Skills · Start here · Agent instructions · Skill library · Contributing
Compile recorded AI coding-agent usage into an append-only, de-duplicated, priced and account-attributed ledger, then render dashboards, study packages and ledger documents. Preserve observed evidence, estimates, exclusions and attribution uncertainty separately.
The ledger is the evidence member of the CompleteTech LLC skill family. It supplies usage and cost evidence to discovery, delivery and billing workflows; it does not issue invoices or authorize charging a client. Its activation key is ai-usage-ledger, not the repository name.
Onboarding separates a safe fixture demonstration from real source onboarding. BRANDING.md describes presentation and downstream handoffs. The previous full README is preserved byte-for-byte in REFERENCE.md for its catalog and examples; use the current Before You Start and Core Workflow in SKILL.md for consent and operational boundaries. In particular, onboarding does not register a schedule: schedule install requires its own operator consent.
Native Muse usage and MCP reporting provides opt-in metadata collection, transactional replay-safe imports, and separately consented reports through Muse's existing MCP client. The guide includes an apply-ready upstream patch and a deterministic end-to-end example.
| Field | Value |
|---|---|
| Skill key | ai-usage-ledger |
| Version | 1.6.0 |
| Core runtime | python3; Python standard library |
| Optional document libraries | See requirements.txt for PDF, DOCX and preview dependencies |
| Repository | CompleteTech-LLC/ai-usage-ledger-skill |
| Package contract | skill-package.json, schema 1 |
The full GitHub checkout contains the logo and previews. A text-only registry bundle can omit binary assets; the new package validator targets the full checkout. This change does not publish a registry release.
From a full checkout with Python 3.12, the package CI baseline:
python tests/make_fixtures.py
python scripts/run_pipeline.py --manifest examples/manifest.fixtures.jsonThe fixture test should print ALL OK. The example manifest writes under tests/out/pipeline; inspect the artifact paths printed by the pipeline. These commands do not initialize your personal ledger, select remote hosts or install schedules. Re-running can replace earlier test outputs.
For real use, first read the disclosure in SKILL.md, confirm the source and retention scope, then follow its python scripts/ledger.py init, run and status workflow. Do not substitute an unattended init --yes for informed operator choices.
For an initialized ledger, create a report from the stored events in your timezone:
python scripts/ledger.py report --since today --tz America/New_YorkThe HTML report shows the exact time window, latest recorded activity, tokens, calls, models, API-equivalent cost and cache savings using the selected price snapshot. Its accompanying CSV and JSON contain only events and totals from that window. It reads the existing store without rescanning sources, changing saved prices or registering schedules.
To use a reviewed price snapshot for this report:
python scripts/ledger.py report --since 2026-10-03 --until 2026-10-03T16:49:12.602-04:00 --tz America/New_York --pricing templates/pricing.jsonThe bundled snapshot includes explicit Sol 6.1, Sol, Luna and Sonnet 5.5 rates, with sources and verification dates on the updated rows. Older saved snapshots may still price these models through fallbacks. See scoped reports for boundaries, freshness, artifact paths and pricing limitations.
Source: assets/diagrams/workflow.mmd.
flowchart LR
A[Operator disclosure and choices] --> B[Remembered configuration]
B --> C[Selected source scan]
C --> D[Append-only store]
D --> E[Price, attribute and validate]
E --> F[Dashboard, study and documents]
F --> G[Review before downstream use]
Example dashboard · Example study · Onboarding transcript · Full reference
The committed examples are synthetic. Keep new outputs outside committed preview paths. Use the actual ledger.py doc --list, query --presets and archive status interfaces documented in the specialist instructions; package alignment does not add a new ledger CLI or test-suite runner.
Reports remain neutral until the operator explicitly chooses branding. Authorized CompleteTech output uses the existing local logo and the CompleteTech report preset. During real onboarding, the existing --brand-preset completetech option selects that preset. Preserve approved custom settings instead of forcing a company identity on personal reports.
Carry evidence, source coverage, price snapshot, attribution confidence, artifact paths, approved brand and approval status to the next specialist. API-equivalent figures are estimates, not proof of a provider invoice or permission to bill.
Read the complete disclosure in SKILL.md before personal onboarding. The skill retains a durable local ledger and chosen output files. Prompt capture, credential-derived account claims, identifiable accounts, raw-log archiving, other profiles/drives, WSL and named SSH hosts are separate opt-ins. Credential tokens and keys must not be retained or published.
SSH operation is not network-free: for explicitly selected hosts, the existing implementation transfers the scanner and retrieves scan outputs using SSH/SCP. Generated pages are self-contained by default; explicitly enabling external resources can cause the viewer's browser to fetch remote resources. The manifest therefore records operator-selected-hosts, not a blanket local-only permission grant.
Schedules are never installed by these package checks or the fixture demonstration. The existing schedule install requires its separate disclosure and operator confirmation; schedule remove removes a registered schedule. Before external sharing, follow the anonymization and publish-check workflow and review the result; a passing check is not approval to publish.
python -m unittest discover -s tests -p 'test_package_*.py' -v
python scripts/validate_package.py
python scripts/validate_quality.pyThe first two commands use only the standard library and test the checkout contract. The original Quality command requires its documented development tools and runs its existing parser, store, security, diagram and bundle checks. See QUALITY.md and CONTRIBUTING.md. No original test suite is replaced or weakened.
Code, templates and documentation retain the MIT license. Names, logos, seals and other brand assets remain subject to BRAND_ASSETS.md. No private assets or font files are added by this alignment.
Goose session schema 17 (audited 1.54.0 source) is supported through the existing
scanner and goose_roots manifest field. After normal onboarding and root review,
python scripts/ledger.py run --tools goose collects native usage records;
--skip-tools goose disables them for a run. No Goose changes or provider calls
are needed. Existing SQLite/JSON/CSV stores, queries and snapshot-priced reports
are reused. Session totals, legacy headers and exports are not extra billable
calls. Provider attribution may be inferred or unavailable; native cost is not
silently treated as billed USD. See setup, mappings, privacy and validation,
including the private temporary DB/WAL snapshot and supported-version limits.