Skip to main content

human prompts

human prompts

Manage HUMΛN prompts as first-class, versioned artifacts — list what's in the registry, validate local files, preview renders, publish new versions, and roll back when something regresses.

Delegation-gated commands (list, versions, publish, rollback, deprecate) require a profile with the matching prompt:* scopes. Local commands (lint, render, cost) run against files on disk and do not call the API.

Subcommands

Subcommand Description Auth
list List prompts you can read prompt:read
versions <key> Version history for one prompt prompt:read
lint Validate local prompt files Local only
render <key> Preview rendered prompt with variables Local only
cost <key> Estimate token cost for a prompt Local only
publish <key> Publish a prompt version to the registry prompt:publish:{key}
rollback <key> Roll back to a previous version prompt:rollback:{key}
deprecate <key> Mark a version deprecated prompt:deprecate:{key}

human prompts list

List prompts visible to your current delegation.

Option Description
--namespace <namespace> Filter by namespace (e.g. companion.task)
--scope <scope> Filter by scope: core, org, marketplace
--format <format> Output format: table (default) or json
human prompts list
# Prompts (12):
#   • legal.contract-analysis v1.2.0 [org]
#   • companion.task.summarize v3.0.1 [core]

human prompts list --namespace companion.task --scope org --format json

human prompts versions

Show semver history for a single prompt key.

Argument Description
<key> Prompt key (e.g. legal.contract-analysis)
human prompts versions legal.contract-analysis
# Versions of legal.contract-analysis:
#   1.2.0 (current) - 2026-05-01T14:22:00.000Z
#   1.1.0 - 2026-04-12T09:00:00.000Z

human prompts lint

Validate all prompt files under a directory (YAML frontmatter, inputSchema vs {{variables}}).

Option Description
--dir <dir> Prompts directory (default: ./prompts)
human prompts lint --dir ./prompts
# Linting 4 prompt(s)...
#   ✓ core/agents/summarize/system
#   ✗ org/legal/contract-analysis — ERROR: missing version in meta
# 4 prompts, 1 error(s), 0 warning(s)
# Exit code 1 when errors exist

human prompts render

Preview a prompt after variable substitution (reads from disk, no API).

Argument / option Description
<key> Prompt id to render
--var <key=value> Repeat for each template variable
--dir <dir> Prompts directory (default: ./prompts)
human prompts render core/agents/summarize/system \
  --var topic="Q2 revenue" \
  --var audience="executive"
# --- core/agents/summarize/system v1.0.0 (core/canon) ---
# You summarize {{topic}} for {{audience}} ...

human prompts cost

Estimate token usage and per-call cost for a prompt file.

Argument / option Description
<key> Prompt id
--model <model> Model for estimation (default: gpt-4o)
--dir <dir> Prompts directory (default: ./prompts)
human prompts cost core/agents/summarize/system --model gpt-4o
# Token Cost Estimate:
#   Prompt:  core/agents/summarize/system
#   Model:   gpt-4o
#   Tokens:  ~842
#   Cost:    ~$0.002106/call

human prompts publish

Publish the active local prompt to the registry. Requires prompt:publish:{key} on your delegation.

Argument / option Description
<key> Prompt id to publish (must exist under --dir)
--scope <scope> Publication scope: org (default) or marketplace
--dir <dir> Prompts directory (default: ./prompts)

Reads version, content, and meta from the local prompt file, then POSTs the full body the API expects.

human prompts lint --dir ./prompts
human prompts publish legal.contract-analysis --scope org
# ✔ Published: prompt://org/legal.contract-analysis@1.3.0

Maps to POST /v1/prompts/:key/publish with body { version, content, meta, input_schema? }. System prompts must include the Prompt Defense Baseline markers (see human prompts lint). The MCP tool human.prompts.publish accepts the same fields.


human prompts rollback

Point the active version at a previous semver. Requires prompt:rollback:{key}.

Argument / option Description
<key> Prompt key
--to <version> Required. Target semver (e.g. 1.0.0)
human prompts rollback legal.contract-analysis --to 1.0.0
# ✔ Rolled back legal.contract-analysis to 1.0.0

Maps to POST /v1/prompts/:key/rollback with body { target_version }.


human prompts deprecate

Mark a specific version deprecated without deleting history. Requires prompt:deprecate:{key}.

Argument / option Description
<key> Prompt key
--version <version> Required. Version to deprecate
human prompts deprecate legal.contract-analysis --version 1.1.0
# ✔ Deprecated legal.contract-analysis v1.1.0

Examples

Complete dev-to-publish workflow with a safety rollback:

# 1. Validate local prompt files before you touch the registry
human prompts lint --dir ./prompts
# ✓ core/agents/support-triage/system

# 2. Preview with production variables
human prompts render core/agents/support-triage/system \
  --var product="Acme Cloud" \
  --var tier="enterprise"

# 3. Publish to your org scope
human prompts publish core/agents/support-triage/system --scope org
# ✔ Published: prompt://org/core/agents/support-triage/system@1.0.0

# 4. Confirm version history
human prompts versions core/agents/support-triage/system
#   1.0.0 (current) - 2026-05-19T10:00:00.000Z

# 5. Ship 1.1.0, discover a regression in staging
human prompts publish core/agents/support-triage/system --scope org
# ✔ Published: ...@1.1.0

# 6. Roll back active pointer — 1.1.0 remains in history, no longer active
human prompts rollback core/agents/support-triage/system --to 1.0.0
# ✔ Rolled back core/agents/support-triage/system to 1.0.0

See also: Developer guide to ctx.prompts · CLI human prompts.