fix(vetkd): use the current ic-cdk call API for management-canister vetKD calls

[raymondk]

Closes #272.

Problem

The Rust snippets called ic_cdk::caller(), ic_cdk::api::call::call, and ic_cdk::api::call::call_with_payment128.

One correction to the issue: it says the code "likely will not compile against its own declared CDK version." It does compile at the pinned ic-cdk = "0.19" — those functions are deprecated there, not gone. The real breakage is one minor version later.

Verification (compiled, not assumed)

I extracted the skill's "Calling management canister directly" block verbatim and built it against the skill's own pinned deps:

ic-cdk	Result
0.19.0 (the pin)	compiles — 3 deprecation warnings (ic_cdk::caller → msg_caller, call/call_with_payment128 → Call::unbounded_wait())
0.20.2 (current release)	3 hard errors — cannot find 'call' in 'api' ×2, cannot find function 'caller' in crate 'ic_cdk'

Builder API confirmed against ic-cdk-0.19.0/src/call.rs: unbounded_wait (:197), with_arg (:216), with_cycles (:279), candid (:342).

After the change, the same extracted block compiles clean with zero warnings on both 0.19.0 and 0.20.2.

Change

• Call::unbounded_wait(Principal::management_canister(), "vetkd_public_key").with_arg(request).await…​.candid()
• vetkd_derive_key likewise, with the cycles moved to .with_cycles(26_000_000_000)
• ic_cdk::caller() → ic_cdk::api::msg_caller() in both the ic-vetkeys and lower-level snippets
• added use ic_cdk::call::Call;

This is the API canister-security, multi-canister, and internet-identity already use at the same pin, which resolves the cross-skill contradiction in the issue.

The ic-cdk = "0.19" pin is deliberately untouched — the modern API exists at 0.19, so no bump is needed here, and pin drift is tracked in #301.

Eval results

The skill had no eval file; this seeds evaluations/vetkd.json with one adversarial output case and three trigger queries.

node scripts/evaluate-skills.js vetkd

━━━ Adversarial: management canister vetkd_derive_key call with cycles ━━━

  WITH skill: 4/4 passed
    ✅ Builds the call with Call::unbounded_wait(Principal::management_canister(), "vetkd_derive_key")
    ✅ Attaches cycles with .with_cycles(...) using ~26 billion for key_1
    ✅ Gets the caller with ic_cdk::api::msg_caller() and captures it before the await
    ✅ Does NOT use ic_cdk::api::call::call, call_with_payment128, or ic_cdk::caller() — all removed in ic-cdk 0.20

  WITHOUT skill: 0/4 passed
    ❌ Builds the call with Call::unbounded_wait(...) → uses ic_cdk::api::call::call_with_payment128 instead
    ❌ Attaches cycles with .with_cycles(...) → passes cycles as a positional arg to call_with_payment128
    ❌ Gets the caller with ic_cdk::api::msg_caller() → calls the deprecated ic_cdk::caller()
    ❌ Does NOT use the removed APIs → uses both call_with_payment128 and ic_cdk::caller()

━━━ Trigger Evals ━━━
  Should trigger: 2/2 correct
  Should NOT trigger: 1/1 correct

  Summary: WITH 4/4 | WITHOUT 0/4 | triggers 2/2 and 1/1

The baseline writes exactly the code that fails to compile on current ic-cdk.

npm run validate — 26 skills validated, all passed (15 warnings, all pre-existing).

Note

PR #162 also touches skills/vetkd/SKILL.md (replacing it with a vetkeys skill). It has been open since April; this change is confined to two snippets, so it should be easy to carry across if #162 lands later — but flagging it in case you'd rather fix this there.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NEm1Tbkypzi8SuXhrMfBek

[github-actions]

Skill Validation Report

Validating skill: /home/runner/work/icskills/icskills/skills/vetkd

Structure

• Pass: SKILL.md found

Frontmatter

• Pass: name: "vetkd" (valid)
• Pass: description: (354 chars)
• Pass: license: "Apache-2.0"
• Pass: compatibility: (16 chars)
• Pass: metadata: (2 entries)

Tokens

• Warning: SKILL.md body is 6819 tokens (spec recommends < 5000)
• Warning: SKILL.md body is 621 lines (spec recommends < 500)

Markdown

• Pass: no unclosed code fences found

Tokens

File	Tokens
SKILL.md body	6,819
Total	6,819

Content Analysis

Metric	Value
Word count	3,428
Code block ratio	0.42
Imperative ratio	0.05
Information density	0.24
Instruction specificity	0.72
Sections	20
List items	29
Code blocks	15

Contamination Analysis

Metric	Value
Contamination level	medium
Contamination score	0.35
Primary language category	systems
Scope breadth	4

• Warning: Language mismatch: config, javascript, shell (3 categories differ from primary)

Result: 2 warnings

Project Checks

✓ Project checks passed for 1 skills (0 warnings)