Skip to content

t1k:ax

FieldValue
Modulet1k-engineering
Version2.10.23
Effortlow
Tools—

Keywords: agent-experience, agent-readable, ax, cli, exit-codes, hook, json-output, mcp, no-prompt, review-for-agents

/t1k:ax
[checklist <type> | audit <path...>]

A tool is agent-ready when a model can run it, parse it and recover from its errors without a human. This skill is the checklist you read before writing it and the audit you run after. It cites, never restates: docs/development-principles.md § “AI-Native by Default” (the rule), docs/marker-namespaces.md (the [t1k:ax-review ...] frame), docs/ai-driven-design.md (scripts emit facts, the model decides) and docs/one-click-handoff.md (when the tool must hand a step to a human).

You areModeDo
About to write or change a CLI, hook, script, skill, agent, rule, MCP tool or docCHECKLISTRead the matching headings of references/ax-checklist.md, build to them, state the ids in the plan AX: line
Reviewing a diff or a hook frame asked you toAUDITRun bun .claude/skills/t1k-ax/scripts/ax-audit.cjs <path...> --json (or --changed), then judge per references/audit-rubric.md

The principle ids, titles, strengths and applicability are read from data/ax-principles.json at audit time; this file never lists them (single source).

Quick list by artifact (ids only; full text in the checklist)

Section titled “Quick list by artifact (ids only; full text in the checklist)”
TypeStart with
cliAX-01, AX-04, AX-07, AX-08, AX-10, AX-14, AX-23
hookAX-04, AX-07, AX-08, AX-14, AX-15, AX-16
skillAX-10, AX-12, AX-16, AX-17, AX-20
agentAX-05, AX-06, AX-09, AX-12
mcpAX-15, AX-21, AX-03, AX-05, AX-16
docAX-19, AX-20, AX-23

Gated (the only ids CI may enforce): AX-04, AX-07, AX-08, AX-15. AX-10 is detected but advisory (strength A, gate: false): a non-blocking severity: advisory finding, never a failing exit; it is enforced at author time by the checklists, the hook nudge and the plan AX: row. Strength B ids are advisory too.

bun .claude/skills/t1k-ax/scripts/ax-audit.cjs <path...> [--type <t>] [--changed [--base <ref>]] [--only AX-NN] [--limit n] --json
ExitMeaningYou do
0no blocking findinganswer each needs-judgment question, one line
1blocking findingapply the finding’s fix, rerun; max 3 attempts, then escalate to a human
2needs input (path, --type, flag)supply the flag named in stderr fix

The script reads the file and never executes it. Output stays under 8 KB; omitted and more say how to see the rest. To waive: ax:waive AX-NN reason="<at least 15 characters>" (see the rubric).

SurfaceWhen it firesFix path
Creator checklists (skill, agent, hook creators)while authoringread CHECKLIST for the type
Plan AX: line (t1k-plan, from plan date 261015)a phase adds a rubric pathwrite AX: <ids and how> or AX: n/a because <reason>
Hook frame [t1k:ax-review ...]after a Write/Edit of a new rubric filerun AUDIT on the named path
CI gate validate-ax-new-tooling (release-action)a PR adds JS/TS toolingfix the cited id or add a waiver with a reason

Known traps live in the gotcha ledger skills/t1k-rsi/ledger/gotchas/t1k-ax.md.

The audit is a deterministic script run: collecting its JSON is mechanical, answering each needs-judgment question and deciding a waiver is not. Routing SSOT: docs/orchestration-rules.md § “Task-Type → Agent Routing”.

StepAgent
Run ax-audit.cjs and relay the JSONt1k-data-verifier
Judge needs-judgment findings, write fixes, decide waiversthe invoking session (reasoning tier, not cheap)
  • references/ax-checklist.md: one heading per principle id, mechanical test and sources
  • references/audit-rubric.md: status semantics, questions per judged id, waiver and retry rules
  • docs/t1k-ax.md: human companion page and source register