Method
/use-skillsThe first skill is choosing when a method applies.
The whole series in one sentence
A library is only load-bearing if you lean on it
A team can document excellent methods: clarify consequential choices, write a plan, create a behavioural test, diagnose before changing and verify before claiming completion. The missing piece is often not another procedure. It is recognising which procedure applies to the task now in front of you.
I call that skill activation. A skill is a reusable method. Activation is the trigger that selects it at the right moment. In a well-designed harness, the trigger is visible and testable rather than dependent on a person or model remembering a document under pressure.
Route the task before granting implementation authority
Visible task signals select the smallest applicable method, including no added method when the risk is low.
- Read task signals Consequences, ambiguity, reversibility, systems and unresolved choices.
- Select a route Plan, test, debug, verify, clarify or none. Inspectable decision
- Require the artifact For example, an approved mini-spec or a failing invariant test.
- Escalate consequential ambiguity A human chooses when authority or product intent is unresolved.
- Review routing evidence False positives and missed activations improve the trigger.
The skill is recognition plus routing, not remembering every document.
A worked routing example
Different tasks should summon different methods
Suppose the request is “fix the misspelling in the footer.” It is specific, reversible and easy to inspect. A short edit and a page check may be enough. Now change the request to “let customers merge duplicate accounts.” That sentence hides identity rules, ownership, billing history, audit records, rollback and data-loss risk. It should activate structured clarification, then planning, test-first implementation and independent verification. The team first decides which identity owns the combined account and records that choice in an approved mini-spec. It then writes an invariant test proving that balances, access and audit history survive the merge. Independent evidence from the combined workflow, including the rollback case, decides whether the change is ready for review.
The difference is not confidence. It is consequence and ambiguity. A useful trigger card might say: if a task changes stored customer records, identity relationships, permissions or financial state, spans several systems or contains an unresolved product choice, stop before implementation and select the relevant method. The card makes the pause operational instead of moral.
Do not ask whether the task feels easy. Ask which consequence or unknown should choose the method.
The pause is the skill
The value is in the instinct that summons the document
The first pause asks three questions. What kind of work is this? What could go wrong if our first interpretation is wrong? Which existing method reduces that risk without adding more ceremony than the task deserves? The answer may be “none.” A reflex that invokes every procedure for every typo is not discipline; it is latency.
This is why writing a skill once is necessary but insufficient. Each method also needs a trigger, exclusions and an expected artifact. “Use for ambiguous changes that affect customer data” is a trigger. “Skip for copy-only edits” is an exclusion. “End with an approved mini-spec” is the artifact that shows the method actually ran.
Build the reflex, not just the library
Wire in the question you'd never think to ask
In an AI workflow, add a routing step before implementation authority. It reads the task, identifies trigger conditions, selects the smallest applicable method and records why. High-consequence uncertainty can require human choice. A mistaken route should be correctable, and teams should review false positives and missed activations as evidence for improving the trigger.
This is not proof that automatic routing will always choose well. Natural language is ambiguous, procedures overlap and novel work may fit no existing skill. The harness should make the choice inspectable, not pretend selection has become objective.
The capstone
The library becomes useful when the work is routed
The professional test is whether the method appears before the consequence, not after the incident. The student test is whether missed and unnecessary activations improve the routing rule. Keep the library small enough to understand, give every skill a clear trigger and let risk choose the ceremony. The first skill is not remembering every document. It is recognising when the work should stop and ask for one.