dl-agent

Agent-facing command line tools, instructions, examples, and sample assets for DriftLedger.

dl-agent is the distribution layer that lets shell-capable agents use DriftLedger without screen scraping the web console. It packages the public CLI, install script, agent instruction templates, request body examples, synthetic demo assets, skills, and future MCP assets in one repository.

The first package is @driftledger/cli, exposed as both:

dl: short command for agents and day-to-day shell use.
driftledger: long command for compatibility and public documentation.

What You Can Do

Install the DriftLedger CLI from a hosted bash script or local checkout.
Connect Codex, Claude Code, OpenClaw, CI jobs, and generic shell agents.
Create or select a workspace.
Upload raw CSV table exports or assembled JSONL reconciliation data.
Create metadata, source bindings, reconciliation models, rules, and execution tasks.
Inspect execution result indexes and incidents through JSON output.
Configure email or webhook alert channels and inspect delivery logs after a run.
Download and reuse synthetic demo assets for training, validation, and incident smoke tests.

Product Vocabulary

Use these names consistently in agent instructions, website copy, command help, and examples:

Term	Meaning
Workspace	Isolation boundary for metadata, data, reconciliation models, rules, runs, incidents, and one compiled RuleForest.
Table metadata	Registered source table shape, ownership, risk level, and descriptive context.
Field metadata	Registered field shape, optional algorithm type hints, join-key flags, primary-key flags, and risk level. Omit type hints unless you need to constrain inference.
Raw table data	CSV exports that preserve the original source-table identity.
Assembled data	JSONL records that already join related source rows into reconciliation samples.
Reconciliation model	The configured model for one business scenario. It defines table relationships, assembly scope, and the parent object for rules. The CLI command group is `check-model` for backend compatibility.
Rule	Human-reviewed invariant attached to a reconciliation model. Rules may come from training, manual creation, or migration.
RuleForest	Workspace-level compiled rule artifact loaded by execution for faster rule evaluation.
Incident	Exception event produced by a run, including rule, evidence rows, and handling state.
Alert channel	Workspace-level email or webhook route used to notify users when a run opens incidents.
Alert delivery	Delivery log for test alerts and incident notifications, optionally filtered by execution task.

The MVP supports two data paths:

Raw table data -> metadata -> source binding -> assembly -> assembled data -> reconciliation model -> rules -> RuleForest -> alert channel -> run -> incidents -> alert delivery
Assembled data -------------------------------------------------------------------------------> reconciliation model -> rules -> RuleForest -> alert channel -> run -> incidents -> alert delivery

Install

Hosted install:

curl -fsSL https://driftledger.fatclaw.com/install.sh | bash
dl config set --api-url https://driftledger.fatclaw.com
dl auth login --email you@example.com --password "<password>"

npm install:

npm install -g @driftledger/cli
dl doctor

Local development install:

git clone https://github.com/tryanswer/dl-agent.git
cd dl-agent
npm install
npm install -g ./packages/cli
dl config set --api-url http://localhost:8088
dl doctor

Agent Setup

Generate the instruction file for the agent runtime in the current project:

dl agent init codex --out AGENTS.md
dl agent init claude --out CLAUDE.md
dl agent init openclaw --out OPENCLAW.md
dl agent init generic --out AGENT.md

Hosted or sandboxed agents can avoid local config writes:

export DRIFTLEDGER_API_URL="https://driftledger.fatclaw.com"
export DRIFTLEDGER_TOKEN="<jwt>"
export DRIFTLEDGER_WORKSPACE_ID="Default"

Configuration precedence is:

CLI flags > environment variables > ~/.driftledger/config.json > Default

If no workspace is provided by --workspace, DRIFTLEDGER_WORKSPACE_ID, or local config, dl uses Default. Pass --workspace <spId> only when an agent should operate outside the default workspace.

Skill Setup

The CLI is the runtime interface. Skills are optional workflow packages that teach an agent how to use the CLI without rediscovering the sequence each time.

Install the DriftLedger skills for Codex:

mkdir -p ~/.codex/skills
cp -R skills/driftledger-cli ~/.codex/skills/
cp -R skills/driftledger-incident-review ~/.codex/skills/

Install the same skills for Claude Code:

mkdir -p ~/.claude/skills
cp -R skills/driftledger-cli ~/.claude/skills/
cp -R skills/driftledger-incident-review ~/.claude/skills/

For OpenClaw or a generic shell agent, keep the skills in the repository and generate a project instruction file:

dl agent init openclaw --out OPENCLAW.md
dl agent init generic --out AGENT.md

Use skills/driftledger-cli for installation, workspace, metadata, data, model, rule, RuleForest, alert, run, and incident commands. Use skills/driftledger-incident-review after a run creates incidents and delivery logs.

How To Use DriftLedger With An Agent

This is the recommended step-by-step sequence for Codex, Claude Code, OpenClaw, CI, and other shell-capable agents.

1. Install and verify the CLI

command -v dl >/dev/null || npm install -g @driftledger/cli
dl doctor

For local development from this repository:

npm install
npm install -g ./packages/cli
dl config set --api-url http://localhost:8088
dl doctor

2. Configure API, authentication, and optional workspace override

dl config set --api-url https://driftledger.fatclaw.com
dl auth login --email you@example.com --password "<password>"
dl auth verify
dl workspace list
dl workspace create --name "Default"

Default is used automatically when no workspace is specified. Use dl workspace activate --workspace <spId> or --workspace <spId> only when the user chooses a different workspace.

Hosted agents can use environment variables instead of local config:

export DRIFTLEDGER_API_URL="https://driftledger.fatclaw.com"
export DRIFTLEDGER_TOKEN="<jwt>"
export DRIFTLEDGER_WORKSPACE_ID="Default"

3. Generate the agent instruction file

dl agent init codex --out AGENTS.md
dl agent init claude --out CLAUDE.md
dl agent init openclaw --out OPENCLAW.md
dl agent init generic --out AGENT.md

Commit the instruction file only after confirming it contains no tokens, passwords, cookies, raw account numbers, webhook secrets, or company data.

4. Download demo assets or choose the data path

The npm CLI package does not bundle sample files. For the public merchant-payment-escrow scenario, download the synthetic assets with the CLI:

dl demo pull
# Default root:
# ~/.driftledger/samples/merchant-payment-escrow-reconciliation

# Optional explicit project-local copy:
dl demo pull --out ./driftledger-demo

The command prints JSON with root, downloaded files, and ready-to-adapt upload commands. Agents should read root from that output when possible.

Use the assembled-data path when another job has already produced joined JSONL:

DEMO_ROOT="${DRIFTLEDGER_DEMO_DIR:-$HOME/.driftledger/samples/merchant-payment-escrow-reconciliation}"
dl dataset create-assembled --display-name merchant-payment-escrow
dl dataset upload-assembled --dataset <datasetId> --file "$DEMO_ROOT/datasets/train.jsonl"

Use the raw-table path when the user uploads CSV exports and DriftLedger should preserve table identity before assembly:

dl metadata col-types
dl metadata upsert --body-file examples/body-files/meta.json
dl data-source upsert --display-name "Payment Order CSV" --type CSV_UPLOAD
dl source-binding upsert --body-file examples/body-files/binding.json
dl dataset create-raw --display-name payment-order --binding-id <bindingId>
dl dataset upload --dataset <datasetId> --file payment_order.csv
dl assembly submit --body-file examples/body-files/assembly.json
dl assembly run --task <assemblyTaskId>

Field types in metadata are optional. If an agent provides them, use only values returned by dl metadata col-types, not SQL or file-parser types such as STRING, DECIMAL, or DATETIME.

5. Create the reconciliation model

The reconciliation model defines table relationships, assembly scope, and the parent object for rules.

dl check-model create --body-file examples/body-files/check-model.json
dl check-model deploy --id <riskModelId>
dl check-model enable --id <riskModelId>

6. Train or add rules

Use training when sample data is available:

dl infer-task submit --body-file examples/body-files/infer-task.json
dl infer-task progress --task <inferTaskId>

Add a reviewed rule before execution:

dl rule types
dl rule add --body-file examples/body-files/rule.json
dl rule list

Manual rule payloads must use a ruleType returned by dl rule types.

7. Build RuleForest

Rule changes should be compiled into the workspace RuleForest before execution.

dl rule-forest build
dl rule-forest status

8. Configure alert delivery

MVP runs should have at least one alert channel before production execution.

dl alerts upsert --body-file examples/body-files/alert-email-channel.json
dl alerts upsert --body-file examples/body-files/alert-webhook-channel.json
dl alerts list
dl alerts test --channel <channelId>

9. Run checks and inspect incidents

dl run submit --body-file examples/body-files/run.json
dl run run --task <taskId>
dl run indexes --task <taskId>
dl incidents task --task <taskId>
dl alerts deliveries --task <taskId>

10. Review and hand off

Use skills/driftledger-incident-review to summarize rule failures, evidence rows, alert delivery state, owner, likely cause, and next action without exposing raw company identifiers.

MVP Workflow

Start with workspace and authentication:

dl doctor
dl auth verify
dl workspace list
# Default is used automatically when no workspace is specified.
# Override only when operating outside Default:
dl config set --workspace <spId>

The commands below use the downloadable merchant-payment-escrow reconciliation demo. It is the fastest path for an agent to prove the complete DriftLedger flow: upload assembled data, create a reconciliation model, add rules, execute validation, and inspect incidents.

Use the assembled-data path when another job already produced joined reconciliation rows:

dl demo pull
DEMO_ROOT="${DRIFTLEDGER_DEMO_DIR:-$HOME/.driftledger/samples/merchant-payment-escrow-reconciliation}"
dl dataset create-assembled --display-name merchant-payment-escrow
dl dataset upload-assembled --dataset <datasetId> --file "$DEMO_ROOT/datasets/train.jsonl"
dl dataset create-assembled --display-name merchant-payment-escrow-anomaly
dl dataset upload-assembled --dataset <anomalyDatasetId> --file "$DEMO_ROOT/datasets/test-with-anomaly.jsonl"
dl check-model create --body-file examples/body-files/check-model.json
# Replace <riskModelCode>, <riskModelVersion>, <riskModelId>, <trainingAssembledDatasetId>,
# and <anomalyAssembledDatasetId>
# in examples/body-files/*.json with the IDs returned by the previous commands.
dl infer-task submit --body-file examples/body-files/infer-task.json
dl infer-task progress --task <inferTaskId>
dl rule types
dl rule add --body-file examples/body-files/rule.json
dl rule-forest build
dl rule-forest status
dl alerts upsert --body-file examples/body-files/alert-email-channel.json
dl alerts test --channel <channelId>
dl run submit --body-file examples/body-files/run.json
dl run run --task <taskId>
dl incidents task --task <taskId>
dl alerts deliveries --task <taskId>

Use the raw-table path when DriftLedger should preserve source-table identity before assembly:

dl metadata col-types
dl metadata upsert --body-file examples/body-files/meta.json
dl data-source upsert --display-name "Payment Order CSV" --type CSV_UPLOAD
dl source-binding upsert --body-file examples/body-files/binding.json
dl dataset create-raw --display-name payment-order --binding-id <bindingId>
dl dataset upload --dataset <datasetId> --file payment_order.csv
dl assembly submit --body-file examples/body-files/assembly.json
dl assembly run --task <assemblyTaskId>

Demo Scenario

samples/merchant-payment-escrow-reconciliation is fully synthetic and models a merchant order paid through an acquiring flow, held in escrow, released after buyer confirmation, cleared, and settled to the merchant. DriftLedger assembles those source rows into one reconciliation record and checks that linkage keys, success statuses, fees, and money amounts remain consistent. The training data also includes payment failures, buyer cancellations, pending buyer confirmation, and delayed clearing so trained rules can learn meaningful preconditions.

Business flow and table relationship design:

flowchart LR
  subgraph ORDER["1. Merchant order"]
    MO["driftledger.merchant_order<br/>order_no<br/>merchant_id<br/>order_status"]
  end

  subgraph PAY["2. Payment"]
    PO["driftledger.payment_order<br/>payment_id<br/>order_no<br/>payment_status<br/>payment_amount, fee_amount, net_amount"]
  end

  subgraph ACQUIRE["3. Acquiring capture"]
    ACQ["driftledger.acquiring_transaction<br/>payment_id<br/>acquiring_status<br/>capture_amount, fee_amount, net_amount"]
  end

  subgraph ESCROW["4. Escrow hold and release"]
    ESC["driftledger.escrow_ledger<br/>escrow_id<br/>payment_id<br/>escrow_status<br/>escrow_amount"]
    REL["driftledger.escrow_release<br/>escrow_id<br/>release_status<br/>release_amount"]
  end

  subgraph SETTLE["5. Clearing and settlement"]
    CLR["driftledger.clearing_record<br/>payment_id<br/>settlement_batch_no<br/>merchant_id<br/>clearing_status<br/>fee_amount, net_amount"]
    SET["driftledger.settlement_record<br/>settlement_batch_no<br/>merchant_id<br/>settlement_status<br/>settlement_amount"]
  end

  MO -- "order_no" --> PO
  PO -- "payment_id" --> ACQ
  PO -- "payment_id" --> ESC
  ESC -- "escrow_id" --> REL
  PO -- "payment_id" --> CLR
  CLR -- "settlement_batch_no + merchant_id" --> SET

The same assembled record carries all seven tables in tableDataMap. Join keys are preserved even in anomaly records, so incidents are caused by checked field values rather than broken assembly.

Scenario coverage:

Scenario profile	Path covered	Rule-learning purpose
`CARD_GOODS_ESCROW_RELEASED`	Card payment succeeds, capture succeeds, escrow is released, clearing and settlement complete.	Positive success-chain rows for amount, fee, and settlement consistency.
`WALLET_DIGITAL_SERVICE_RELEASED`	Wallet payment succeeds with instant buyer confirmation.	Same success-chain invariants across a different payment method.
`BANK_TRANSFER_B2B_ESCROW_RELEASED`	Bank-transfer B2B payment succeeds and releases after acceptance.	Larger B2B-style amounts with the same settlement invariants.
`CARD_AUTHORIZATION_FAILED`	Card authorization fails before capture.	Teaches rules to require `payment_status = 'PAID'` and `acquiring_status = 'CAPTURED'`.
`BUYER_CANCELLED_BEFORE_CAPTURE`	Buyer cancels before payment capture.	Prevents rules from treating canceled records as capture or settlement mismatches.
`ESCROW_AWAITING_BUYER_CONFIRMATION`	Payment is captured and held, but release is still pending.	Teaches release and settlement rules to require release/clearing preconditions.
`CLEARING_DELAYED_AFTER_RELEASE`	Escrow is released, but clearing and settlement are still pending.	Teaches settlement rules to require `clearing_status = 'CLEARED'` and `settlement_status = 'SETTLED'`.

Model identity:

Sample asset model ID: DL_SYNTH_MERCHANT_ESCROW_001
Sample asset version: synthetic-v1
Backend risk model code: returned by dl check-model create; use it as <riskModelCode> in rule payloads.
Trigger table: driftledger.payment_order
Trigger filter: driftledger.payment_order.payment_status == 'PAID'
Data shape: assembled JSONL records with tableDataMap entries for all joined tables.

Data split:

File	Records	Purpose
`datasets/train.jsonl`	160	Clean assembled records with success, failed, canceled, pending, and delayed status flows for rule training.
`datasets/test.jsonl`	24	Clean validation set with the same status mix.
`datasets/test-with-anomaly.jsonl`	4	Anomaly-only validation records with controlled success-chain amount mismatches.

The anomaly fixture is intentionally small. It keeps all join keys intact and injects mismatches into acquiring capture amount, escrow release amount, clearing fee amount, and merchant settlement amount. These should produce incidents when success-chain rules are active.

Generated Rule Candidates

These representative rules are the demo rules an agent should create or expect from training on this fixture:

Rule	What it protects
`driftledger.payment_order#payment_amount = driftledger.acquiring_transaction#capture_amount WHERE driftledger.payment_order#payment_status = 'PAID' AND driftledger.acquiring_transaction#acquiring_status = 'CAPTURED'`	A paid and captured order must be captured for the same amount.
`driftledger.payment_order#payment_amount = driftledger.escrow_ledger#escrow_amount WHERE driftledger.payment_order#payment_status = 'PAID' AND driftledger.escrow_ledger#escrow_status = 'HELD'`	A paid order held in escrow must hold the paid amount.
`driftledger.escrow_ledger#escrow_amount = driftledger.escrow_release#release_amount WHERE driftledger.escrow_ledger#escrow_status = 'HELD' AND driftledger.escrow_release#release_status = 'RELEASED'`	Buyer-confirmed release must release the held amount.
`driftledger.payment_order#fee_amount = driftledger.clearing_record#fee_amount WHERE driftledger.payment_order#payment_status = 'PAID' AND driftledger.clearing_record#clearing_status = 'CLEARED'`	Clearing fee must match the payment fee.
`driftledger.clearing_record#net_amount = driftledger.settlement_record#settlement_amount WHERE driftledger.clearing_record#clearing_status = 'CLEARED' AND driftledger.settlement_record#settlement_status = 'SETTLED'`	Merchant settlement must pay the cleared net amount.

The checked-in examples/body-files/rule.json uses the settlement rule because datasets/test-with-anomaly.jsonl is built to fail it deterministically.

Sample Assets

Installed CLI users should download the public synthetic demo assets:

dl demo pull
dl demo pull --out ./driftledger-demo
dl demo pull --force

By default, dl demo pull writes to:

~/.driftledger/samples/merchant-payment-escrow-reconciliation

Use --out <dir> for a project-local copy, --force to refresh existing files, and --source-base <url> or DRIFTLEDGER_DEMO_BASE_URL for a mirror. The command prints JSON with root, each file status, and upload command templates.

The source assets in this repository live under:

samples/merchant-payment-escrow-reconciliation

Files:

datasets/train.jsonl: clean assembled records for rule training.
datasets/test.jsonl: clean assembled records for validation.
datasets/test-with-anomaly.jsonl: four anomaly-only records for incident verification.
models/demo_model.jsonl: synthetic reconciliation model for the same scenario.
manifest.json: fixture metadata, scenario name, counts, and privacy notes.
README.md: dataset-local scenario notes and privacy guard details.

Privacy guard:

npm run verify:samples

Regenerate deterministic synthetic samples:

npm run sync:samples

sync:samples generates the records locally and writes to samples/merchant-payment-escrow-reconciliation; it does not read company fixtures or sibling repositories.

Repository Layout

The repository is organized by distribution surface. Keep CLI runtime code, agent instructions, workflow docs, request bodies, synthetic samples, release scripts, skill packages, and future MCP wrappers in separate directories so agents can update one surface without touching the others.

dl-agent/
  package.json                   npm workspace scripts, package metadata, and Node.js engine contract
  package-lock.json              reproducible dependency lockfile
  .env.example                   local CLI/API configuration example, no secrets
  .editorconfig                  editor defaults for portable formatting
  .gitignore                     generated files and local environment exclusions
  .github/
    workflows/ci.yml             CI for CLI tests and sample privacy guard
    ISSUE_TEMPLATE/              bug and feature templates
      bug_report.md
      feature_request.md
    pull_request_template.md      PR checklist
  agents/
    AGENTS.md                    Codex instruction contract
    CLAUDE.md                    Claude Code instruction contract
    OPENCLAW.md                  OpenClaw instruction contract
    AGENT.md                     generic shell-agent instruction contract
  assets/
    logo.svg                     README and website brand asset
  docs/
    architecture.md              maintainer architecture notes
    commands.md                  command groups, aliases, and backend endpoints
    publishing.md                npm, install-script, and GitHub release notes
    workflows.md                 raw-table and assembled-data MVP flows
  examples/
    body-files/
      meta.json                  table and field metadata request body
      binding.json               source-column binding request body
      assembly.json              raw-table assembly task request body
      check-model.json           reconciliation model request body
      infer-task.json            rule training task request body
      rule.json                  confirmed rule request body
      run.json                   execution task request body
      alert-email-channel.json   email notification channel request body
      alert-webhook-channel.json webhook notification channel request body
  packages/
    cli/                         npm package published as @driftledger/cli
      package.json               package name, binary aliases, scripts, and publish metadata
      bin/driftledger.js         executable entrypoint
      src/cli.js                 command execution and HTTP transport
      src/core.js                argument parsing, config precedence, endpoint mapping
      test/core.test.js          CLI contract tests
      README.md                  package-level usage notes
  scripts/
    install.sh                   hosted install script
    sync-demo-assets.mjs         generate deterministic synthetic demo assets
    verify-demo-assets.mjs       privacy guard for sample assets
    README.md                    script-specific maintenance notes
  samples/
    merchant-payment-escrow-reconciliation/
      README.md                  scenario notes and privacy guard explanation
      datasets/
        train.jsonl              clean assembled records for training
        test.jsonl               clean assembled records for validation
        test-with-anomaly.jsonl  four anomaly-only records for incident verification
      models/
        demo_model.jsonl         synthetic reconciliation model
      manifest.json              sample metadata and privacy notes
  skills/
    README.md                    DriftLedger skill packaging surface
    driftledger-cli/             CLI-first workflow skill package
      SKILL.md                   install, auth, workspace, metadata, alerts, run workflow
      examples/                  reusable command transcripts and body files
      scripts/                   optional validators or setup helpers
    driftledger-incident-review/ incident triage and remediation skill package
      SKILL.md                   incident summarization, alert delivery, and remediation workflow
      templates/                 report and handoff templates
  mcp/                           future MCP server packaging surface
    README.md                    MCP transport and tool contract notes
  CHANGELOG.md                   release history
  CODE_OF_CONDUCT.md             contributor behavior expectations
  CONTRIBUTING.md                contribution workflow
  LICENSE                        Apache-2.0 license text
  SECURITY.md                    security reporting and data handling guidance

Directory responsibilities:

Directory	Owner surface	Change when
`packages/cli`	`dl` / `driftledger` command runtime	Adding commands, changing config precedence, or updating API mappings.
`agents`	Codex, Claude Code, OpenClaw, and generic agent instruction presets	Changing how agents should call the CLI or handle secrets.
`examples/body-files`	Reviewable JSON payload templates	Backend request contracts change, or the MVP workflow needs a new request body.
`samples/merchant-payment-escrow-reconciliation`	Synthetic model, train/test data, and anomaly fixture	Refreshing demo data, model structure, or privacy guard metadata.
`scripts`	Install, synthetic sample generation, and privacy verification automation	Publishing install scripts or regenerating public samples.
`skills`	Agent skill packages	Packaging DriftLedger workflows as reusable Codex/Claude/OpenClaw skills.
`mcp`	Future MCP server or manifest packages	Exposing DriftLedger atoms as MCP tools instead of shell commands.
`docs`	Maintainer architecture and publishing notes	Explaining package boundaries, release steps, or operational decisions.

Placeholder directories such as skills/driftledger-cli and mcp are committed early so future work has a stable target instead of adding ad-hoc top-level folders.

Extension Points

Add new surfaces in the narrowest directory that owns the change:

Need	Put it in
A new backend atom exposed to agents	`packages/cli/src/core.js`, `packages/cli/src/cli.js`, and `packages/cli/test/core.test.js`
A reusable request body	`examples/body-files`
A new sample scenario	`samples/<semantic-scenario-name>`
A privacy or data-shape guard	`scripts/verify-*.mjs` plus CI
A Codex/Claude/OpenClaw instruction template	`agents` now, then `skills/<skill-name>` when packaged
A future MCP transport	`mcp`

Documentation Map

Start here when installing or validating the full agent workflow: README.md.
Use packages/cli/README.md for package-level CLI usage.
Use docs/architecture.md to understand package boundaries and command contracts.
Use docs/commands.md to inspect command groups, aliases, and backend endpoints.
Use docs/workflows.md for raw-table and assembled-data MVP flows.
Use docs/publishing.md before publishing the npm package or hosted install script.
Use samples/merchant-payment-escrow-reconciliation/README.md to understand the demo data, model, and anomaly fixture.
Use scripts/README.md when changing install, sync, or privacy verification scripts.
Use skills/README.md for Codex/Claude/OpenClaw workflow packaging boundaries.

Command Surface

Core commands:

dl doctor
dl version
dl config get
dl config set --api-url <url> --token <jwt>
dl config set --workspace <spId>
dl auth login --email <email> --password <password>
dl auth verify
dl auth refresh

Workspace, metadata, and data:

dl workspace list
dl workspace create --name "Default"
dl metadata col-types
dl metadata upsert --body-file meta.json
dl data-source upsert --display-name "Payment Order CSV" --type CSV_UPLOAD
dl source-binding upsert --body-file binding.json
dl dataset create-raw --display-name payment-order --binding-id <bindingId>
dl dataset upload --dataset <datasetId> --file payment_order.csv
dl assembly submit --body-file assembly.json
dl assembly run --task <assemblyTaskId>
dl dataset create-assembled --display-name assembled
dl dataset upload-assembled --dataset <datasetId> --file assembled.jsonl

Models, rules, runs, and incidents:

dl check-model create --body-file check-model.json
dl check-model deploy --id <riskModelId>
dl check-model enable --id <riskModelId>
dl infer-task submit --body-file infer-task.json
dl infer-task status --task <inferTaskId>
dl infer-task progress --task <inferTaskId>
dl rule types
dl rule add --body-file rule.json
dl rule list
dl rule-forest build
dl rule-forest status
dl run submit --body-file run.json
dl run run --task <taskId>
dl run indexes --task <taskId>
dl incidents list
dl incidents task --task <taskId>

Alerts and delivery logs:

dl alerts upsert --body-file examples/body-files/alert-email-channel.json
dl alerts upsert --body-file examples/body-files/alert-webhook-channel.json
dl alerts list
dl alerts enable --channel <channelId>
dl alerts disable --channel <channelId>
dl alerts test --channel <channelId>
dl alerts deliveries
dl alerts deliveries --task <taskId>

CLI Contract

Normal command responses are JSON on stdout.
Errors are structured JSON on stderr with ok:false.
Complex payloads should be stored in files and passed through --body-file.
Tokens should be provided through DRIFTLEDGER_TOKEN or local config, not prompts.
The MVP supports CSV for raw table uploads and JSONL for assembled reconciliation data.

Development

npm install
npm test
npm run verify:samples
npm run check
npm run cli -- doctor

Run the CLI package tests directly:

npm --workspace @driftledger/cli test

Publishing

Dry-run package publication:

npm run check
npm --workspace @driftledger/cli pack --dry-run

The hosted installer should be served at:

https://driftledger.fatclaw.com/install.sh

Security

Do not commit tokens, passwords, cookies, or raw company datasets.
Do not commit real alert recipients, webhook URLs, or webhook secrets.
Run npm run verify:samples before publishing sample changes.
Keep user credentials out of generated agent instruction files.
Prefer environment variables for CI and hosted agents.

License

Apache-2.0

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
.github		.github
agents		agents
assets		assets
docs		docs
examples/body-files		examples/body-files
mcp		mcp
packages/cli		packages/cli
samples/merchant-payment-escrow-reconciliation		samples/merchant-payment-escrow-reconciliation
scripts		scripts
skills		skills
.editorconfig		.editorconfig
.env.example		.env.example
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
package-lock.json		package-lock.json
package.json		package.json

Folders and files

Latest commit

History

Repository files navigation

dl-agent

What You Can Do

Product Vocabulary

Install

Agent Setup

Skill Setup

How To Use DriftLedger With An Agent

1. Install and verify the CLI

2. Configure API, authentication, and optional workspace override

3. Generate the agent instruction file

4. Download demo assets or choose the data path

5. Create the reconciliation model

6. Train or add rules

7. Build RuleForest

8. Configure alert delivery

9. Run checks and inspect incidents

10. Review and hand off

MVP Workflow

Demo Scenario

Generated Rule Candidates

Sample Assets

Repository Layout

Extension Points

Documentation Map

Command Surface

CLI Contract

Development

Publishing

Security

License

About

Topics

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages