Tool calling na actions

Day 12 of 30 · AI Jeneretivu 2026: Jenga App na Agent za AI

One-liner: Unganisha model na tools na ulinde actions zisizo salama.
Time: 20 hadi 30 dakika
Deliverable: Tool Schema na Mfano wa Action

Learning goal

You will be able to: Kufafanua tool schema na kuendesha tool call salama.

Success criteria (observable)

Tool schema ina name, inputs, na output shape.
Tool call inaendeshwa kwa mfano halisi.
Actions zisizo salama zimezuiwa.

Output you will produce

Deliverable: Tool Schema na Mfano wa Action
Format: JSON schema na test log
Where saved: Repo na noti za kozi ndani ya folda ya kozi

Who

Primary persona: Digital nomad anayejenga AI features zenye tools Secondary persona(s): Watumiaji wanaochochea actions Stakeholders (optional): Washirika wa ujenzi

What

What it is

Ufafanuzi wa tool unaoonyesha model nini inaweza kuita na vipi. Mfano mdogo unaothibitisha tool call inafanya kazi kama ilivyokusudiwa.

What it is not

Si mfumo kamili wa automation au permissions. Ni hatua ya kwanza ya actions salama na zinazotabirika.

2-minute theory

Tool calling huleta AI zaidi ya maandishi.
Schema wazi hupunguza makosa ya input.
Guardrails huzuia actions zisizotarajiwa.

Key terms

Tool schema: Ufafanuzi wa inputs na outputs kwa tool.
Guardrail: Sheria ya kuzuia action zisizo salama.

Where

Applies in

Agent workflows
Backend services

Does not apply in

Uzalishaji wa maudhui pekee bila actions

Touchpoints

Tool definitions
Action logs
Permission checks

When

Use it when

Unataka AI ichochee action
Unahitaji inputs zinazotabirika

Frequency

Kila unapoongeza tool mpya

Late signals

Tool calls kushindwa kwa inputs mbaya
Actions zisizotarajiwa kutokea

Why it matters

Practical benefits

AI feature zinakuwa na maana zaidi
Failures chache kwenye automation
Usalama na uaminifu bora

Risks of ignoring

Workflows kuvunjika
Actions zisizo salama

Expectations

Improves: uthabiti na usalama
Does not guarantee: maamuzi kamili

How

Step-by-step method

Fafanua jina la tool na kusudi lake.
Bainisha input fields na types.
Fafanua output shape.
Ongeza guardrails kwa actions hatari.
Endesha test call na rekodi matokeo.

Do and don't

Validate inputs kabla ya utekelezaji
Rekodi kila action

Don't

Kuruhusu tools ziende bila checks
Kufungua actions hatari bila control

Common mistakes and fixes

Mistake: Schema legevu. Fix: Ongeza required fields na types.
Mistake: Hakuna guardrails. Fix: Zuia parameters hatari.

Done when

Tool schema imefafanuliwa na kupimwa.
Guardrails zinazuia inputs hatari.
Logs zinaonyesha call iliyofanikiwa.

Guided exercise (10 to 15 min)

Inputs

Wazo la tool moja
Mfano wa input values

Steps

Andika tool schema.
Ongeza guardrails kwa inputs hatari.
Endesha test call na rekodi output.

Output format

Field	Value
Tool name
Input schema
Guardrails
Test result

Pro tip: Ikiwa tool inaweza kubadilisha data, ongeza hatua ya kuthibitisha.

Independent exercise (5 to 10 min)

Task

Ongeza guardrail moja zaidi na urudie test.

Output

Schema iliyoboreshwa na test log mpya.

Self-check (yes/no)

Je, schema iko wazi na ina types?
Je, actions zisizo salama zimezuiwa?
Je, test call imeandikwa?
Je, logs zimehifadhiwa?

Baseline metric (recommended)

Score: Tool call 1 imefanikiwa na guardrails
Date: 2026-02-07
Tool used: Notes app

Bibliography (sources used)

OpenAI Tools Guide. OpenAI. 2026-02-06. Read: https://platform.openai.com/docs/guides/tools
OWASP API Security Top 10. OWASP. 2024-01-01. Read: https://owasp.org/www-project-api-security/