// Generated from the parser

CLI

Every subcommand of the Halbert command-line interface, generated from the parser in Halbert/main.py. Help text, flags, defaults and required arguments below are the parser’s own, not a retelling of them.

This file is generated. Edit Halbert/main.py (or scripts/gen_cli_reference.py for the prose around the tables), then run:

arch -arm64 .venv/bin/python scripts/gen_cli_reference.py

--check reports drift without writing, which is the form for CI:

arch -arm64 .venv/bin/python scripts/gen_cli_reference.py --check

On Apple Silicon the arch -arm64 prefix is required: the venv’s python3 is a universal2 binary and otherwise starts its x86_64 slice, which cannot load the project’s compiled extensions.

Invoking the CLI

python Halbert/main.py <command> [options]

The installed halbert console script does not reach this CLI — it starts the web dashboard. Run the CLI through python Halbert/main.py until the entry point is rewired, which is tracked as README-1 in ROADMAP.md §3.

Global flags, accepted before the subcommand:

Flag	Description
`-h`, `--help`	Show help and exit. Also accepted after any subcommand.
`--version`	Print the version and the GPLv3 notice, then exit.

A subcommand is required; running the CLI with no arguments is an error.

The GPLv3 notice printed by --version tells you to type halbert license. Until the console script is rewired, read that as python Halbert/main.py license.

Commands

67 subcommands.

Information and legal — Print what this build is, what it is licensed under, and what is in the tree.

Command	Description
`info`	Show product info and legal summary
`license`	Show software license and third-party notices
`roadmap`	Show Phase 1 roadmap (docs/Phase1/ROADMAP.md)
`show`	Show a project file (relative to Halbert/)
`show-doc`	Show a docs file (relative to docs/)

Ingestion — Follow a live source and append telemetry events to JSONL.

Command	Description
`ingest-journald`	Run journald follower and write JSONL (uses config/ingestion.yml by default)
`ingest-hwmon`	Poll hwmon temp sensors and write JSONL (uses config/ingestion.yml by default)

Configuration tracking — Snapshot the configuration files under watch, and diff or index those snapshots.

Command	Description
`snapshot-configs`	Snapshot tracked config files (uses config/config-registry.yml by default)
`watch-configs`	Watch tracked config files and snapshot on changes
`diff-configs`	Diff latest vs previous config snapshots (or provide –prev/–curr)
`index-configs`	Index canonical config records into the vector DB

Knowledge and retrieval — Query what the machine knows about itself and about Linux administration.

Command	Description
`index-query`	Query the vector index (self_knowledge_all)
`ask`	Ask a Linux/DevOps question using RAG + LLM (Phase 7)
`eval-golden`	Run the golden task harness (stub)

RAG corpus — Curate the retrieval corpus: add sources, inspect trust tiers, rebuild.

Command	Description
`rag-add`	Add URL to RAG corpus (Phase 10)
`rag-check`	Check URL before adding (Phase 10)
`rag-bulk`	Add multiple URLs from file (Phase 10)
`rag-sitemap`	Discover URLs from sitemap (Phase 10)
`rag-sources`	List RAG sources and trust tiers (Phase 10)
`rag-stats`	Show RAG corpus statistics (Phase 10)
`rag-merge`	Rebuild RAG corpus from all sources (Phase 10)

Runtime — Drive the runtime engine by hand, one iteration at a time.

Command	Description
`runtime-tick`	Run one iteration of the runtime (LangGraph if available, else fallback)

Policy — Show the loaded policy and evaluate what it would decide for a tool call.

Command	Description
`policy-show`	Show the currently loaded policy (from config_dir/policy.yml or defaults)
`policy-eval`	Evaluate policy decision for a tool with inputs JSON file (apply path)

Integrity and ledgers — Verify the append-only ledgers and reproject their derived views.

Command	Description
`audit-verify`	Check the audit log for tampering (edits, truncation, deleted shards)
`vault-rebuild`	Reproject the change ledger into the Markdown vault (generated; overwrites edits)
`consent-verify`	Check the consent ledger for tampering (chain + projection)
`consent-rebuild`	Reproject consent-state.json from the authoritative ledger (generated; refuses a tampered chain)

Scheduler — The job queue: add work, list it, cancel it.

Command	Description
`scheduler-add`	Add a job to the scheduler queue (Phase 2)
`scheduler-list`	List scheduler jobs (Phase 2)
`scheduler-cancel`	Cancel a pending job (Phase 2)

Autonomous executor — Cron-shaped autonomous jobs and the one-shot autonomous task runner.

Command	Description
`executor-status`	Show autonomous executor status (Phase 3 M3)
`executor-schedule`	Schedule a cron job (Phase 3 M3)
`executor-list`	List scheduled jobs (Phase 3 M3)
`executor-cancel`	Cancel a scheduled job (Phase 3 M3)
`autonomous-run`	Run an autonomous task with LLM decision making (Phase 3 M3)

Approval — The human-in-the-loop queue an autonomous action waits in.

Command	Description
`approval-list`	List pending approval requests (Phase 3 M4)
`approval-history`	Show approval history (Phase 3 M4)

Autonomy guardrails — Pause and resume autonomy, and read what the guardrails caught.

Command	Description
`autonomy-status`	Show autonomy guardrail status (Phase 3 M6)
`autonomy-pause`	Pause autonomous operations (safe-mode) (Phase 3 M6)
`autonomy-resume`	Resume autonomous operations (Phase 3 M6)
`autonomy-anomalies`	View detected anomalies (Phase 3 M6)
`autonomy-recovery`	View recovery action history (Phase 3 M6)

Personas, memory and context — Which persona is speaking, what it remembers, and how context is detected.

Command	Description
`persona-list`	List available personas (Phase 4 M1)
`persona-status`	Show active persona status (Phase 4 M1)
`persona-switch`	Switch to a different persona (Phase 4 M1)
`memory-purge`	Purge persona-specific memory (Phase 4 M1)
`memory-export`	Export persona memory to file (Phase 4 M1)
`context-detect`	Detect current user context (Phase 4 M4)
`context-prefs`	Show/update context detection preferences (Phase 4 M4)

Models and prompts — Connection slots, routing, system prompts, and training-data preparation.

Command	Description
`model-status`	Show model status (Phase 3)
`model-test`	Test model with a prompt (Phase 3)
`model-list-all`	List all available models (Phase 5 M1)
`model-router-status`	Show model router status (Phase 5 M1)
`model-select-specialist`	Select specialist model (Phase 5 M1)
`model-disable-specialist`	Disable specialist (orchestrator-only) (Phase 5 M1)
`prompt-show`	Show system prompt for a mode (Phase 3)
`prompt-init`	Initialize default prompt configuration (Phase 3)
`prepare-training-data`	Prepare training data for persona (Phase 5 M4)

Hardware and configuration — What this machine can carry, and the wizard that writes the answer down.

Command	Description
`hardware-detect`	Detect hardware and show the model size budget (Phase 5 M3)
`config-wizard`	Run configuration wizard (Phase 5 M3)
`config-validate`	Validate model configuration (Phase 5 M3)

Performance — Latency and throughput metrics, and the alerts raised from them.

Command	Description
`performance-status`	Show model performance metrics (Phase 5 M5)
`performance-alerts`	Show performance alerts (Phase 5 M5)
`performance-reset`	Reset performance metrics (Phase 5 M5)

Dashboard and artifacts — Serve the web dashboard, or build the JSON artifacts it reads.

Command	Description
`dashboard`	Start Halbert web dashboard (Phase 3 M5)
`build-dashboard`	Build dashboard JSON artifacts under data/dashboard

Information and legal

`info`

Show product info and legal summary

python Halbert/main.py info

`license`

Show software license and third-party notices

python Halbert/main.py license [--full] [--third-party]

Argument	Description	Default
`--full`	Show complete verbatim GNU GPLv3 license text.	off
`--third-party`	Show third-party notices (RAG corpus sources, dependencies, foundation models)	off

`roadmap`

Show Phase 1 roadmap (docs/Phase1/ROADMAP.md)

python Halbert/main.py roadmap

This command cannot find its file. It reads <repo>/docs/Phase1/ROADMAP.md, but docs/ is gitignored (.gitignore:108) and carries no Phase1/ directory, so a clone has nothing for it to open. It prints File not found and still exits 0. The live roadmap is ROADMAP.md at the repository root.

`show`

Show a project file (relative to Halbert/)

python Halbert/main.py show path

Argument	Description	Default
`path` (positional)	Relative path, e.g., planning/phase-1-alpha.md.	required

`show-doc`

Show a docs file (relative to docs/)

python Halbert/main.py show-doc path

Argument	Description	Default
`path` (positional)	Relative path, e.g., Phase1/ROADMAP.md.	required

Paths resolve under the same gitignored docs/ tree, not under documentation/. Whatever is there is machine-local and absent from a fresh clone. Read a file in documentation/ directly, or use show, which resolves under Halbert/.

Ingestion

`ingest-journald`

Run journald follower and write JSONL (uses config/ingestion.yml by default)

python Halbert/main.py ingest-journald [--config CONFIG] [--schema SCHEMA]

Argument	Description	Default
`--config CONFIG`	Path to ingestion.yml.	—
`--schema SCHEMA`	Path to telemetry-event.schema.json.	—

`ingest-hwmon`

Poll hwmon temp sensors and write JSONL (uses config/ingestion.yml by default)

python Halbert/main.py ingest-hwmon [--config CONFIG]

Argument	Description	Default
`--config CONFIG`	Path to ingestion.yml.	—

Configuration tracking

`snapshot-configs`

Snapshot tracked config files (uses config/config-registry.yml by default)

python Halbert/main.py snapshot-configs [--manifest MANIFEST]

Argument	Description	Default
`--manifest MANIFEST`	Path to config-registry.yml.	—

`watch-configs`

Watch tracked config files and snapshot on changes

python Halbert/main.py watch-configs [--manifest MANIFEST] [--interval INTERVAL]

Argument	Description	Default
`--manifest MANIFEST`	Path to config-registry.yml.	—
`--interval INTERVAL`	Polling interval seconds when watchdog is unavailable (default 600)	`600`

`diff-configs`

Diff latest vs previous config snapshots (or provide –prev/–curr)

python Halbert/main.py diff-configs [--prev PREV] [--curr CURR]

Argument	Description	Default
`--prev PREV`	Path to previous snapshot JSON.	—
`--curr CURR`	Path to current snapshot JSON.	—

`index-configs`

Index canonical config records into the vector DB

python Halbert/main.py index-configs

Knowledge and retrieval

`index-query`

Query the vector index (self_knowledge_all)

python Halbert/main.py index-query [-k K] text

Argument	Description	Default
`text` (positional)	Query text.	required
`-k K`	Top-K.	`5`

`ask`

Ask a Linux/DevOps question using RAG + LLM (Phase 7)

python Halbert/main.py ask [--model MODEL] [--no-llm] [--top-k TOP_K] [question ...]

Argument	Description	Default
`question ...` (positional)	Question to ask (can be multiple words)	optional
`--model MODEL`	Ollama model to use (default: the model configured in Settings → AI Models)	—
`--no-llm`	Just retrieve docs, no LLM generation.	off
`--top-k TOP_K`	Number of documents to retrieve (default: 3) (int)	`3`

`eval-golden`

Run the golden task harness (stub)

python Halbert/main.py eval-golden

RAG corpus

`rag-add`

Add URL to RAG corpus (Phase 10)

python Halbert/main.py rag-add [--trust] url

Argument	Description	Default
`url` (positional)	URL to add.	required
`--trust`	Trust unknown source.	off

`rag-check`

Check URL before adding (Phase 10)

python Halbert/main.py rag-check url

Argument	Description	Default
`url` (positional)	URL to check.	required

`rag-bulk`

Add multiple URLs from file (Phase 10)

python Halbert/main.py rag-bulk [--trust] file

Argument	Description	Default
`file` (positional)	File with URLs (one per line)	required
`--trust`	Trust unknown sources.	off

`rag-sitemap`

Discover URLs from sitemap (Phase 10)

python Halbert/main.py rag-sitemap [--include [INCLUDE ...]] [--exclude [EXCLUDE ...]]
    [--max MAX] [--add] [--trust]
    sitemap_url

Argument	Description	Default
`sitemap_url` (positional)	Sitemap URL (e.g., https://site.com/sitemap.xml)	required
`--include [INCLUDE ...]`	Only include URLs containing these patterns.	—
`--exclude [EXCLUDE ...]`	Exclude URLs containing these patterns.	—
`--max MAX`	Max URLs to show (default: 50) (int)	`50`
`--add`	Actually add the URLs (otherwise just preview)	off
`--trust`	Trust unknown sources.	off

`rag-sources`

List RAG sources and trust tiers (Phase 10)

python Halbert/main.py rag-sources

`rag-stats`

Show RAG corpus statistics (Phase 10)

python Halbert/main.py rag-stats

`rag-merge`

Rebuild RAG corpus from all sources (Phase 10)

python Halbert/main.py rag-merge

Runtime

`runtime-tick`

Run one iteration of the runtime (LangGraph if available, else fallback)

python Halbert/main.py runtime-tick

Policy

`policy-show`

Show the currently loaded policy (from config_dir/policy.yml or defaults)

python Halbert/main.py policy-show

`policy-eval`

Evaluate policy decision for a tool with inputs JSON file (apply path)

python Halbert/main.py policy-eval --tool TOOL --inputs INPUTS

Argument	Description	Default
`--tool TOOL`	Tool name (e.g., write_config, schedule_cron)	required
`--inputs INPUTS`	Path to JSON file with tool inputs.	required

Integrity and ledgers

`audit-verify`

Check the audit log for tampering (edits, truncation, deleted shards)

python Halbert/main.py audit-verify [--dir DIR] [--json]

Argument	Description	Default
`--dir DIR`	Audit log directory (default: <log_dir>/audit)	—
`--json`	Emit machine-readable JSON.	off

`vault-rebuild`

Reproject the change ledger into the Markdown vault (generated; overwrites edits)

python Halbert/main.py vault-rebuild [--json]

Argument	Description	Default
`--json`	Emit machine-readable JSON.	off

`consent-verify`

Check the consent ledger for tampering (chain + projection)

python Halbert/main.py consent-verify [--data-dir DATA_DIR] [--config-dir CONFIG_DIR]
    [--json]

Argument	Description	Default
`--data-dir DATA_DIR`	Data dir (default: the resolved data dir)	—
`--config-dir CONFIG_DIR`	Config dir (default: the resolved config dir)	—
`--json`	Emit machine-readable JSON.	off

`consent-rebuild`

Reproject consent-state.json from the authoritative ledger (generated; refuses a tampered chain)

python Halbert/main.py consent-rebuild [--data-dir DATA_DIR] [--config-dir CONFIG_DIR]
    [--json]

Argument	Description	Default
`--data-dir DATA_DIR`	Data dir (default: the resolved data dir)	—
`--config-dir CONFIG_DIR`	Config dir (default: the resolved config dir)	—
`--json`	Emit machine-readable JSON.	off

Scheduler

`scheduler-add`

Add a job to the scheduler queue (Phase 2)

python Halbert/main.py scheduler-add --id ID --task TASK --schedule SCHEDULE
    [--priority PRIORITY] --inputs INPUTS

Argument	Description	Default
`--id ID`	Job ID.	required
`--task TASK`	Task name (e.g., snapshot_configs)	required
`--schedule SCHEDULE`	Cron expression or ISO timestamp.	required
`--priority PRIORITY`	Priority (1=highest, 10=lowest)	`5`
`--inputs INPUTS`	Path to JSON file with task inputs.	required

`scheduler-list`

List scheduler jobs (Phase 2)

python Halbert/main.py scheduler-list [--state STATE]

Argument	Description	Default
`--state STATE`	Filter by state (pending, running, completed, failed, cancelled)	—

`scheduler-cancel`

Cancel a pending job (Phase 2)

python Halbert/main.py scheduler-cancel --id ID

Argument	Description	Default
`--id ID`	Job ID.	required

Autonomous executor

`executor-status`

Show autonomous executor status (Phase 3 M3)

python Halbert/main.py executor-status

`executor-schedule`

Schedule a cron job (Phase 3 M3)

python Halbert/main.py executor-schedule --job-id JOB_ID --cron CRON
    [--description DESCRIPTION] [--daemon]

Argument	Description	Default
`--job-id JOB_ID`	Unique job identifier.	required
`--cron CRON`	Cron expression JSON (e.g., ‘{“hour”: 2, “minute”: 0}’)	required
`--description DESCRIPTION`	Job description.	—
`--daemon`	Run in daemon mode (keep running)	off

`executor-list`

List scheduled jobs (Phase 3 M3)

python Halbert/main.py executor-list

`executor-cancel`

Cancel a scheduled job (Phase 3 M3)

python Halbert/main.py executor-cancel --job-id JOB_ID

Argument	Description	Default
`--job-id JOB_ID`	Job ID to cancel.	required

`autonomous-run`

Run an autonomous task with LLM decision making (Phase 3 M3)

python Halbert/main.py autonomous-run --task-type {health_check,log_cleanup}
    [--confidence CONFIDENCE] [--context CONTEXT]
    [--no-approval]

Argument	Description	Default
`--task-type TASK_TYPE`	Type of autonomous task. Choices: `health_check`, `log_cleanup`.	required
`--confidence CONFIDENCE`	Confidence threshold (default: 0.7) (float)	—
`--context CONTEXT`	Task context as JSON string.	—
`--no-approval`	Skip approval prompts (auto-approve)	off

Approval

`approval-list`

List pending approval requests (Phase 3 M4)

python Halbert/main.py approval-list

`approval-history`

Show approval history (Phase 3 M4)

python Halbert/main.py approval-history [--limit LIMIT] [--approved-only]

Argument	Description	Default
`--limit LIMIT`	Max records to show. (int)	`10`
`--approved-only`	Only show approved requests.	off

Autonomy guardrails

`autonomy-status`

Show autonomy guardrail status (Phase 3 M6)

python Halbert/main.py autonomy-status

`autonomy-pause`

Pause autonomous operations (safe-mode) (Phase 3 M6)

python Halbert/main.py autonomy-pause --reason REASON

Argument	Description	Default
`--reason REASON`	Reason for pausing autonomy.	required

`autonomy-resume`

Resume autonomous operations (Phase 3 M6)

python Halbert/main.py autonomy-resume --user USER

Argument	Description	Default
`--user USER`	Your username (for audit trail)	required

`autonomy-anomalies`

View detected anomalies (Phase 3 M6)

python Halbert/main.py autonomy-anomalies [--hours HOURS]

Argument	Description	Default
`--hours HOURS`	Look back hours (default: 24) (int)	`24`

`autonomy-recovery`

View recovery action history (Phase 3 M6)

python Halbert/main.py autonomy-recovery [--limit LIMIT]

Argument	Description	Default
`--limit LIMIT`	Max records to show (default: 20) (int)	`20`

Personas, memory and context

`persona-list`

List available personas (Phase 4 M1)

python Halbert/main.py persona-list

`persona-status`

Show active persona status (Phase 4 M1)

python Halbert/main.py persona-status

`persona-switch`

Switch to a different persona (Phase 4 M1)

python Halbert/main.py persona-switch --to {it_admin,friend,custom} [--user USER]
    [--confirm]

Argument	Description	Default
`--to TO`	Target persona. Choices: `it_admin`, `friend`, `custom`.	required
`--user USER`	Your username (for audit trail)	—
`--confirm`	Confirm the switch (required to execute)	off

`memory-purge`

Purge persona-specific memory (Phase 4 M1)

python Halbert/main.py memory-purge --persona PERSONA [--user USER] [--confirm]
    [--no-export]

Argument	Description	Default
`--persona PERSONA`	Persona to purge (e.g., friend, custom)	required
`--user USER`	Your username (for audit trail)	—
`--confirm`	Confirm the purge (required to execute)	off
`--no-export`	Skip export before purge (not recommended)	off

`memory-export`

Export persona memory to file (Phase 4 M1)

python Halbert/main.py memory-export --persona PERSONA --output OUTPUT

Argument	Description	Default
`--persona PERSONA`	Persona to export (e.g., friend, custom)	required
`--output OUTPUT`	Output JSONL file path.	required

`context-detect`

Detect current user context (Phase 4 M4)

python Halbert/main.py context-detect

`context-prefs`

Show/update context detection preferences (Phase 4 M4)

python Halbert/main.py context-prefs [--enabled] [--disable] [--auto-switch]
    [--no-auto-switch]

Argument	Description	Default
`--enabled`	Enable context detection.	off
`--disable`	Disable context detection.	off
`--auto-switch`	Enable auto-switching (no confirmation)	off
`--no-auto-switch`	Disable auto-switching (suggest only)	off

Models and prompts

`model-status`

Show model status (Phase 3)

python Halbert/main.py model-status

`model-test`

Test model with a prompt (Phase 3)

python Halbert/main.py model-test --prompt PROMPT [--max-tokens MAX_TOKENS]

Argument	Description	Default
`--prompt PROMPT`	Test prompt.	required
`--max-tokens MAX_TOKENS`	Max tokens to generate. (int)	—

`model-list-all`

List all available models (Phase 5 M1)

python Halbert/main.py model-list-all

`model-router-status`

Show model router status (Phase 5 M1)

python Halbert/main.py model-router-status

`model-select-specialist`

Select specialist model (Phase 5 M1)

python Halbert/main.py model-select-specialist --model MODEL [--provider PROVIDER]

Argument	Description	Default
`--model MODEL`	Model ID as served by your endpoint.	required
`--provider PROVIDER`	Provider (ollama, llamacpp, mlx)	`ollama`

`model-disable-specialist`

Disable specialist (orchestrator-only) (Phase 5 M1)

python Halbert/main.py model-disable-specialist

`prompt-show`

Show system prompt for a mode (Phase 3)

python Halbert/main.py prompt-show [--mode MODE] [--context CONTEXT]

Argument	Description	Default
`--mode MODE`	Prompt mode (interactive, autonomous, it_admin, friend, custom)	`interactive`
`--context CONTEXT`	Additional task context.	—

`prompt-init`

Initialize default prompt configuration (Phase 3)

python Halbert/main.py prompt-init

`prepare-training-data`

Prepare training data for persona (Phase 5 M4)

python Halbert/main.py prepare-training-data [--persona PERSONA] [--output OUTPUT]
    [--num-synthetic NUM_SYNTHETIC]
    [--list-personas] [--validate VALIDATE]

Argument	Description	Default
`--persona PERSONA`	Persona name.	—
`--output OUTPUT`	Output file path (JSONL)	—
`--num-synthetic NUM_SYNTHETIC`	Number of synthetic samples (default: 10) (int)	`10`
`--list-personas`	List available persona templates.	off
`--validate VALIDATE`	Validate existing training data file.	—

Hardware and configuration

`hardware-detect`

Detect hardware and show the model size budget (Phase 5 M3)

python Halbert/main.py hardware-detect

`config-wizard`

Run configuration wizard (Phase 5 M3)

python Halbert/main.py config-wizard [--auto] [--model MODEL] [--peer PEER]

Argument	Description	Default
`--auto`	Run automatically without prompts.	off
`--model MODEL`	Guide model ID as served by your endpoint (default: largest installed model that fits)	—
`--peer PEER`	Compute peer address (hostname:port) for offload-only devices (<4GB RAM): all LLM work is offloaded to the peer.	—

`config-validate`

Validate model configuration (Phase 5 M3)

python Halbert/main.py config-validate

Performance

`performance-status`

Show model performance metrics (Phase 5 M5)

python Halbert/main.py performance-status

`performance-alerts`

Show performance alerts (Phase 5 M5)

python Halbert/main.py performance-alerts [--severity {info,warning,error,critical}]
    [--hours HOURS]

Argument	Description	Default
`--severity SEVERITY`	Filter by severity. Choices: `info`, `warning`, `error`, `critical`.	—
`--hours HOURS`	Show alerts from last N hours (default: 24) (int)	`24`

`performance-reset`

Reset performance metrics (Phase 5 M5)

python Halbert/main.py performance-reset [--model MODEL]

Argument	Description	Default
`--model MODEL`	Reset specific model (or all if not specified)	—

Dashboard and artifacts

`dashboard`

Start Halbert web dashboard (Phase 3 M5)

python Halbert/main.py dashboard [--host HOST] [--port PORT]

Argument	Description	Default
`--host HOST`	Host to bind to.	`127.0.0.1`
`--port PORT`	Port to bind to. (int)	`8000`

`build-dashboard`

Build dashboard JSON artifacts under data/dashboard

python Halbert/main.py build-dashboard

Exit codes

Code	Meaning
0	Success
1	The command reported a failure
2	argparse rejected the arguments (unknown command, missing required flag)

A handler that returns an integer sets the exit code; one that returns None exits 0. Commands that only print an error message without returning a status still exit 0 — check the output, not just $?.

Environment variables

Variable	Description
`HALBERT_CONFIG_DIR`	Override the configuration directory
`HALBERT_DATA_DIR`	Override the data directory
`HALBERT_LOG_DIR`	Override the log directory
`OLLAMA_HOST`	Endpoint used when resolving model licence attribution

The mixed-case spellings (Halbert_CONFIG_DIR, Halbert_DATA_DIR, Halbert_LOG_DIR) are still read as fallbacks. Prefer the uppercase names.

Examples

Track a configuration change end to end:

python Halbert/main.py snapshot-configs
# edit /etc/ssh/sshd_config
python Halbert/main.py snapshot-configs
python Halbert/main.py diff-configs

Ask a question against the retrieval corpus, without a model in the loop:

python Halbert/main.py ask "Why did docker.service fail to start?"
python Halbert/main.py ask --no-llm --top-k 5 "journald disk usage"

Check that the ledgers have not been tampered with, machine-readably:

python Halbert/main.py audit-verify --json
python Halbert/main.py consent-verify --json

Schedule a nightly job and leave the executor running:

python Halbert/main.py executor-schedule \
  --job-id nightly-health \
  --cron '{"hour": 2, "minute": 0}' \
  --description "Nightly system health check" \
  --daemon

Export a persona’s memory before purging it:

python Halbert/main.py memory-export --persona friend --output /tmp/friend.jsonl
python Halbert/main.py memory-purge --persona friend --confirm