Free to read · no account needed

9 min · skills

Invoke skills explicitly vs letting them auto-trigger

Goal: Know when you must call by name (/command or click) vs when precise auto-triggers are safe—with anti-triggers written.

Best for: Beginners with a small skills library who feel surprised by which skill fired.

Skills are reusable instruction packs. Auto-invoke can save clicks and also apply the wrong pack. Explicit beats surprise until your library is small and well-named.

Explicit invoke is training wheels that keep trust. Auto-invoke only after triggers are precise—and after you have watched it fire correctly more than once.

Why beginners skip this

Full auto-invoke feels like fewer clicks—a misfire runs the wrong skill. Explicit invoke first, then auto.

Step 1 — Explicit pattern

On a real artifact, call the skill by exact name; keep auto off for a week.

Use skill: [skill-name]
Task: [ ]
Extra boundaries for this run: [ ]

Step 2 — When auto is OK

  • One obvious skill matches the task
  • Skill has a clear description and tests
  • Blast radius is low (draft-only)
Example

Filled prompt

Rewrite this skill trigger from vague auto to explicit: currently fires on any message containing "help". New behavior: user must type /support-reply or click the skill. Keep the skill body for calm 120-word support drafts.

What good output looks like

  • Skill purpose is one job for “Prefer explicit skill calls while learning—so you ” — not a company brain
  • Done + golden example present and tested
  • Never-send / never-delete honored
  • Pass/fail: fail-case run once (skills-invoke-explicit-vs-auto #1)

Step 3 — Debug wrong auto

If the wrong skill ran: rename skills more distinctly, improve descriptions, or disable auto and stay explicit for a week.

Scene: the wrong skill almost fired

Auto almost applied email-polish to your status bullets. You switched to explicit status-bullets-v3 and the shape snapped back. Auto stays off another week.

Naming that prevents collisions

Prefer status-bullets-v3 over helper. Descriptions should state when *not* to use the skill. Ambiguous names invite wrong auto-triggers.

Rewrite my skill name + one-line description to reduce auto misfires. Current: [ ]. Competing skill: [ ].
Example

Filled prompt

Skill "invoice-reminder" could auto-run on any mail mentioning $.
Set explicit-only. Write exact invoke phrase + 2 do-not-auto cases (legal threads, VIP).
Run one explicit invoke on a sample draft.

What good output looks like

  • Done + golden example present and tested
  • Never-send / never-delete honored
  • Pass/fail: fail-case run once (skills-invoke-explicit-vs-auto #2)
  • Skill purpose is one job for “Prefer explicit skill calls while learning—so you ” — not a company brain

Save this card

Card: explicit invoke until trigger test passes · auto is a privilege. Write the trigger test on the card.

Before you start

Open the skill card (or blank one-pager). Purpose must be one job for “Prefer explicit skill calls while learning—so.” Golden example slot cannot stay empty. Guide: skills-invoke-explicit-vs-auto. Keep the real job visible the whole time; a blank box means stop and fetch the artifact first.

How to use this week

This week keep one real skill on explicit invoke only; turn auto off and note what a misfire would do.

Explicit until boring

Auto-invoke only after triggers are precise and you have watched correct fires repeatedly.

Keep it real

Run this guide’s Practice now on work you will actually ship this week—not a fictional scenario. Dated notes beat perfect theory.

Ship the practice

Close the tab only after Practice now is done on a real artifact.

Do not bookmark-only

Finish Practice now, then tick this guide on your path. Titles unread are not skills.

Close the loop

Write one line: next time I enable full auto with no misfire note, I refuse.

Example

Filled prompt

Skill "invoice-reminder" must not auto-run on legal@ threads.
Explicit invoke phrase + 2 do-not-auto cases. Leave auto off.
Default explicit; include one anti-trigger.

What good output looks like

  • Never-send / never-delete honored
  • Skill purpose is one job for “Prefer explicit skill calls while learning—so you ” — not a company brain
  • Pass/fail: fail-case run once (skills-invoke-explicit-vs-auto #3)
  • Done + golden example present and tested

Red flag

Auto-invoke with a fuzzy trigger; auto as the default on day one—stay explicit until the trigger test passes.

Pitfalls

PitfallFix
Overlapping skill namesRename; one job each
Auto on high-radius skillsExplicit only
No tests before autoGolden example first
Example

Filled prompt

Temptation: skip the habit for skills-invoke-explicit-vs-auto to save five minutes.
Job still: Prefer explicit skill calls while
Smallest correct pass: Done visible · checkable artifact · Unknowns [?].
Boundaries: draft only; no silent send; no invented metrics.
Return: artifact + refuse-line said aloud.

What good output looks like

  • Names the shortcut refused (specific to skills-invoke-explicit-vs-auto)
  • Still delivers checkable output for “Prefer explicit skill calls while”
  • Unknowns labeled; draft-only
  • Pass/fail: refuse-line dated (skills-invoke-explicit-vs-auto #4)

Practice now

Surprise log: one unexpected auto-invoke → disable auto (Asia/Shanghai).

Log invokes: note each auto fire; surprise = flip to explicit (Asia/Shanghai).

Auto-off night: force explicit invoke for one session (Asia/Shanghai).

Invoke drill: mark one skill explicit-only; document why auto would be dangerous (Asia/Shanghai).

Thin-spot drill: rewrite the Save card with real nouns from your work (repo, doc, inbox). Generic placeholders fail the drill.

Tonight switch one skill to explicit invoke, log three misfire reasons; open auto only after whitelist passes.

Reality check: after Practice now, write one sentence you will refuse next time—and keep it on the Save card dated Asia/Shanghai.

One sentence takeaway

Explicit invoke first; auto off; note what a misfire would do.

Word-floor practice

Run the Today card once on a real artifact. Write pass/fail + never-do with date Asia/Shanghai. If you only skimmed, do not mark this guide done.

Next

What is a skill · Test before trust · Anti-patterns

← Back to home · Learn · Tools · Compare · Start