CtrlK
BlogDocsLog inGet started
Tessl Logo

migrate

Migrate an application with hardcoded LLM prompts to a full LaunchDarkly AgentControl implementation in five stages: audit the code, wrap the call, move the tools, add tracking, attach evaluators. Use when the user wants to externalize model/prompt configuration, move from direct provider calls (OpenAI, Anthropic, Bedrock, Gemini, Strands) to a managed config, or stage a full hardcoded-to-LaunchDarkly migration.

70

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

81%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An exceptionally actionable, well-sequenced migration workflow with real verification checkpoints and genuine progressive disclosure into six verified reference files. The main cost is token efficiency: the highest-value rules are each repeated three to four times across the warning block, stages, Edge Cases, and 'What NOT to Do' sections, inflating a ~570-line body that could be materially tightened without losing clarity.

Suggestions

Consolidate the repeated failure-mode rules (load_chat_model deletion, /configs-targeting fallthrough flip, once-per-turn tracker/config lifetime): state each once in its owning stage and reference it from the 'What NOT to Do' section instead of restating the full rationale in all four locations.

Move the Edge Cases table's framework-specific rows (Strands, LangChain model construction, custom StateGraph) into agent-mode-frameworks.md, keeping only a one-line pointer per row in SKILL.md — the detail already exists in the reference file.

Resolve or restructure the ../built-in-metrics/ cross-bundle links: either confirm the sibling skill ships alongside this one, or copy the handful of directly needed snippets (custom extractor, streaming TTFT) into this bundle's references so every link resolves from the skill root.

DimensionReasoningScore

Conciseness

Nearly every sentence carries SDK-specific knowledge Claude does not have (no concept-teaching padding), but the same failure-mode rules are restated three to four times each — the load_chat_model deletion rule appears in the warning block, Stage 2 sub-step 1, the 'What NOT to Do' section, and Edge Cases; the /configs-targeting fallthrough rule and the once-per-turn tracker rule are similarly repeated. This fits the score-3 anchor ('mostly efficient but could be tightened'); it avoids score 2 because the verbosity is redundancy of high-value content, not generic explanation or fluff.

3 / 5

Actionability

Executable Python and Node snippets (SDK init, fallback constructors, tracker wiring, judge guards), exact package names with version floors, before/after diffs, and exact guard idioms ('if judge and judge.enabled') — copy-paste-ready and covering the common cases, matching the score-5 anchor. The few partial snippets (commented agent-mode example) are clearly flagged as shape illustrations with pointers to full worked examples in references.

5 / 5

Workflow Clarity

Five explicitly ordered stages with sequencing rationale ('Wrap before you add tools...'), a hard STOP checkpoint after the read-only audit with four explicit response forms, and a verification sub-step at every stage including fallback-path testing with an invalid SDK key — a clear sequence with explicit validation and error-recovery feedback loops, the score-5 anchor. The hand-off model (prepare inputs, wait for the user to run the sibling skill) is stated unambiguously and reinforced in 'What NOT to Do'.

5 / 5

Progressive Disclosure

All six references/*.md files cited in the body exist in the bundle, are one level deep, are linked with section-level anchors, and are indexed both in a References section and a coverage-tier matrix — good structure per the score-4 anchor. It falls short of score 5 because the 570-line body inlines substantial detail duplicated from the reference files (Edge Cases table, three-failure-mode warnings), and many links point into a sibling skill outside this bundle (../built-in-metrics/...) which cannot be resolved from the bundle itself.

4 / 5

Total

17

/

20

Passed

Description

92%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: third-person, concrete five-stage action list, explicit 'Use when...' triggers with provider names, and a clearly bounded niche. Its only weakness is modest trigger-term coverage — a few natural user phrasings are absent.

DimensionReasoningScore

Specificity

The description enumerates five concrete migration stages — 'audit the code, wrap the call, move the tools, add tracking, attach evaluators' — giving comprehensive, specific coverage of the action surface, matching the score-5 anchor. No gaps in the action list; it is not merely 'several specific actions' with minor gaps (score 4).

5 / 5

Completeness

It explicitly answers both 'what' (the five named stages of the migration) and 'when' via a concrete 'Use when the user wants to...' clause with three distinct trigger conditions — the exact shape of the score-5 anchor. The 'when' is explicit and specific, so it exceeds the score-4 anchor where the trigger clause could be more specific.

5 / 5

Trigger Term Quality

Good natural-term coverage: 'externalize model/prompt configuration', 'move from direct provider calls (OpenAI, Anthropic, Bedrock, Gemini, Strands) to a managed config', and 'hardcoded-to-LaunchDarkly migration' are phrases users would actually say. A few natural variants (e.g. 'move prompts out of code', 'prompt migration') are missing, so it fits the score-4 anchor rather than the comprehensive score-5 anchor.

4 / 5

Distinctiveness Conflict Risk

It names a specific product (LaunchDarkly AgentControl) and migration-specific triggers (externalize config, hardcoded-to-LaunchDarkly migration), forming a clear niche with minimal conflict risk against other skills — the score-5 anchor. It is far more distinct than the 'mostly distinct, minor overlap' score-4 example.

5 / 5

Total

19

/

20

Passed

Validation

87%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

skill_md_line_count

SKILL.md is long (573 lines); consider splitting into references/ and linking

Warning

relative_links

Relative link issues: 10 suspicious

Warning

Total

14

/

16

Passed

Repository
launchdarkly/ai-tooling
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.