Provides 98-rule spec for building AI-agent-safe CLI tools with JSON output, error handling, input validation, safety guardrails, exit codes, and agent self-description.
From antigravity-awesome-skillsnpx claudepluginhub sickn33/antigravity-awesome-skills --plugin antigravity-awesome-skillsThis skill uses the workspace's default tool permissions.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Enables AI agents to execute x402 payments with per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents pay for APIs, services, or other agents.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
When building or modifying CLI tools, follow these rules to make them safe and reliable for AI agents to use.
A comprehensive design specification for building AI-native CLI tools. It defines 98 rules across three certification levels (Agent-Friendly, Agent-Ready, Agent-Native) with prioritized requirements (P0/P1/P2). The spec covers structured JSON output, error handling, input contracts, safety guardrails, exit codes, self-description, and a feedback loop via a built-in issue system.
--humanThis spec uses two orthogonal axes:
core, recommended, ecosystemP0, P1, P2Use layers for migration and certification:
agent/, skills, issue, inline contextCertification maps to layers:
core rules passcore + recommended rules passDefault is agent mode (JSON). Explicit flags to switch:
$ mycli list # default = JSON output (agent mode)
$ mycli list --human # human-friendly: colored, tables, formatted
$ mycli list --agent # explicit agent mode (override config if needed)
Every CLI tool MUST have an agent/ directory at its project root. This is the
tool's identity and behavior contract for AI agents.
agent/
brief.md # One paragraph: who am I, what can I do
rules/ # Behavior constraints (auto-registered)
trigger.md # When should an agent use this tool
workflow.md # Step-by-step usage flow
writeback.md # How to write feedback back
skills/ # Extended capabilities (auto-registered)
getting-started.md
Each level includes all rules from the previous level.
Priority tag [P0]=agent breaks without it, [P1]=agent works but poorly, [P2]=nice to have.
Goal: CLI is a stable, callable API. Agent can invoke, parse, and handle errors.
Output -- default is JSON, stable schema
[P0] O1: Default output is JSON. No --json flag needed[P0] O2: JSON MUST pass jq . validation[P0] O3: JSON schema MUST NOT change within same versionError -- structured, to stderr, never interactive
[P0] E1: Errors -> {"error":true, "code":"...", "message":"...", "suggestion":"..."} to stderr[P0] E4: Error has machine-readable code (e.g. MISSING_REQUIRED)[P0] E5: Error has human-readable message[P0] E7: On error, NEVER enter interactive mode -- exit immediately[P0] E8: Error codes are API contracts -- MUST NOT rename across versionsExit Code -- predictable failure signals
[P0] X3: Parameter/usage errors MUST exit 2[P0] X9: Failures MUST exit non-zero -- never exit 0 then report error in stdoutComposability -- clean pipe semantics
[P0] C1: stdout is for data ONLY[P0] C2: logs, progress, warnings go to stderr ONLYInput -- fail fast on bad input
[P1] I4: Missing required param -> structured error, never interactive prompt[P1] I5: Type mismatch -> exit 2 + structured errorSafety -- protect against agent mistakes
[P1] S1: Destructive ops require --yes confirmation[P1] S4: Reject ../../ path traversal, control charsGuardrails -- runtime input protection
[P1] G1: Unknown flags rejected with exit 2[P1] G2: Detect API key / token patterns in args, reject execution[P1] G3: Reject sensitive file paths (*.env, *.key, *.pem)[P1] G8: Reject shell metacharacters in arguments (; | && $())Goal: CLI is self-describing, well-named, and pipe-friendly. Agent discovers capabilities and chains commands without trial and error.
Self-Description -- agent discovers what CLI can do
[P1] D1: --help outputs structured JSON with commands[][P1] D3: Schema has required fields (help, commands)[P1] D4: All parameters have type declarations[P1] D7: Parameters annotated as required/optional[P1] D9: Every command has a description[P1] D11: --help outputs JSON with help, rules, skills, commands[P1] D15: --brief outputs agent/brief.md content[P1] D16: Default JSON (agent mode), --human for human-friendly[P2] D2/D5/D6/D8/D10: per-command help, enums, defaults, output schema, versionInput -- unambiguous calling convention
[P1] I1: All flags use --long-name format[P1] I2: No positional argument ambiguity[P2] I3/I6/I7: --json-input, boolean --no-X, array paramsError
[P1] E6: Error includes suggestion field[P2] E2/E3: errors to stderr, error JSON validSafety
[P1] S8: --sanitize flag for external input[P2] S2/S3/S5/S6/S7: default deny, --dry-run, no auto-update, destructive markingExit Code
[P1] X1: 0 = success[P2] X2/X4-X8: 1=general, 10=auth, 11=permission, 20=not-found, 30=conflictComposability
[P1] C6: No interactive prompts in pipe mode[P2] C3/C4/C5/C7: pipe-friendly, --quiet, pipe chain, idempotencyNaming -- predictable flag conventions
[P1] N4: Reserved flags (--agent, --human, --brief, --help, --version, --yes, --dry-run, --quiet, --fields)[P2] N1/N2/N3/N5/N6: consistent naming, kebab-case, max 3 levels, --version semverGuardrails
[P1] I8/I9: no implicit state, non-interactive auth[P1] G6/G9: precondition checks, fail-closed[P2] G4/G5/G7: permission levels, PII redaction, batch limits| Flag | Semantics | Notes |
|---|---|---|
--agent | JSON output (default) | Explicit override |
--human | Human-friendly output | Colors, tables, formatted |
--brief | One-paragraph identity | For sync into agent config |
--help | Full self-description JSON | Brief + commands + rules + skills + issue |
--version | Semver version string | |
--yes | Confirm destructive ops | Required for delete/destroy |
--dry-run | Preview without executing | |
--quiet | Suppress stderr output | |
--fields | Filter output fields | Save tokens |
Goal: CLI has identity, behavior contract, skill system, and feedback loop. Agent can learn the tool, extend its use, and report problems -- full closed-loop collaboration.
Agent Directory -- tool identity and behavior contract
[P1] D12: agent/brief.md exists[P1] D13: agent/rules/ has trigger.md, workflow.md, writeback.md[P1] D17: agent/rules/*.md have YAML frontmatter (name, description)[P1] D18: agent/skills/*.md have YAML frontmatter (name, description)[P2] D14: agent/skills/ directory + skills subcommandResponse Structure -- inline context on every call
[P1] R1: Every response includes rules[] (full content from agent/rules/)[P1] R2: Every response includes skills[] (name + description + command)[P1] R3: Every response includes issue (feedback guide)Meta -- project-level integration
[P2] M1: AGENTS.md at project root[P2] M2: Optional MCP tool schema export[P2] M3: CHANGELOG.md marks breaking changesFeedback -- built-in issue system
[P2] F1: issue subcommand (create/list/show)[P2] F2: Structured submission with version/context/exit_code[P2] F3: Categories: bug / requirement / suggestion / bad-output[P2] F4: Issues stored locally, no external service dependency[P2] F5: issue list / issue show <id> queryable[P2] F6: Issues have status tracking (open/in-progress/resolved/closed)[P2] F7: Issue JSON has all required fields (id, type, status, message, created_at, updated_at)[P2] F8: All issues have status field$ mycli list
{"result": [{"id": 1, "title": "Buy milk", "status": "todo"}], "rules": [...], "skills": [...], "issue": "..."}
{
"error": true,
"code": "AUTH_EXPIRED",
"message": "Access token expired 2 hours ago",
"suggestion": "Run 'mycli auth refresh' to get a new token"
}
0 success 10 auth failed 20 resource not found
1 general error 11 permission denied 30 conflict/precondition
2 param/usage error
Implement by layer -- each phase gets you the next certification level.
Phase 1: Agent-Friendly (core)
--json flag needed{ error, code, message, suggestion } to stderr--yes guard on destructive operationsPhase 2: Agent-Ready (+ recommended)
8. --help returns structured JSON (help, commands[], rules[], skills[])
9. --brief reads and outputs agent/brief.md content
10. --human flag switches to human-friendly format
11. Reserved flags: --agent, --version, --dry-run, --quiet, --fields
12. Exit codes: 20 not found, 30 conflict, 10 auth, 11 permission
Phase 3: Agent-Native (+ ecosystem)
13. Create agent/ directory: brief.md, rules/trigger.md, rules/workflow.md, rules/writeback.md
14. Every command response appends: rules[] + skills[] + issue
15. skills subcommand: list all / show one with full content
16. issue subcommand for feedback (create/list/show/close/transition)
17. AGENTS.md at project root
suggestion field in every error responseagent/brief.md to one paragraph for token efficiencyProblem: CLI outputs human-readable text by default, breaking agent parsing
Solution: Make JSON the default output format; add --human flag for human-friendly mode
Problem: Errors reported in stdout with exit code 0 Solution: Always exit non-zero on failure and write structured error JSON to stderr
Problem: CLI prompts for missing input interactively Solution: Return structured error with suggestion field and exit immediately
@cli-best-practices - General CLI design patterns (this skill focuses specifically on AI agent compatibility)