CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

recsys-pipeline-architect

Design composable recommendation, ranking, and feed pipelines using the six-stage Source→Hydrator→Filter→Scorer→Selector→SideEffect framework popularized by xAI's open-sourced X For You algorithm. Use when building any system that picks "the top K items for a (user, context)" — content feeds, search ranking, RAG rerankers, task prioritizers, notification triage, ad selection.

Quality

83%

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

75%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.

A tight, well-structured instruction skill: the six-stage framework, ordered 8-step workflow, and trade-off menus with explicit defaults give Claude genuinely actionable design guidance, and redundancy is minor (Anti-Patterns largely restate Hard Rules). The main gaps are the absence of any scaffold template or interface example in the body — the runnable scaffolds live only in an external upstream repo, not in the skill bundle — and no explicit validation step proving the generated scaffold runs.

Suggestions

Merge the Anti-Patterns section into the Hard Rules (four of its five bullets restate rules 5–7 and the isolation default) and trim the Overview paragraph that restates the frontmatter description, cutting ~10 redundant lines.

Add an explicit validation checkpoint to the workflow, e.g. a step 9: run the generated scaffold's test suite and iterate until green — this operationalizes Hard Rule 5 ('the scaffold must run').

Include at least one minimal runnable scaffold skeleton or the core stage interfaces inline (or as local reference files under references/) so the skill is self-contained without requiring a fetch of the upstream GitHub repo.

DimensionReasoningScore

Conciseness

The body is dense and purposeful — a stage table, an 8-step workflow, three concrete trade-offs with explicit defaults, hard rules with specific numbers (100–300ms latency budget), and a npx install command — with no padding about concepts Claude already knows. Minor trimmable redundancy keeps it below anchor 5: the Overview paragraph restates the description, the Anti-Patterns section re-lists Hard Rules 5–7 almost verbatim ("Pseudocode 'for illustration'", "Synchronous side effects", "Scoring before filtering" vs. rules 5–7), and "Why this exact order" partially re-derives what the table's ordering already implies. Not a 3, since these are minor instances rather than several whole unnecessary sections.

4 / 5

Actionability

The 8-step workflow gives concrete per-step guidance (step 4 enumerates the actual filters: duplicate, self, age, block/mute, previously-served, eligibility; step 5 fixes the scorer chain order; step 6 specifies sort-descending/top-K/stratified-mix) and the trade-offs section presents decision menus with explicit recommended defaults ("Default to isolation", "Recommend multi-action when the user expects to tune frequently"). It falls short of anchor 5 because the core deliverable — the generated scaffold — has no template, interface sketch, or code example in the body; step 8 just says "Generate the scaffold in the user's stack". Not a 3: per the instruction-only-skill scoring note, the absence of code is not penalized where guidance is this specific, and there are no pseudocode stubs posing as code.

4 / 5

Workflow Clarity

The eight steps are clearly sequenced with a built-in clarification checkpoint ("one round, three questions only if missing") and each step names its concrete inputs and outputs. Checkpoint gaps keep it below anchor 5: there is no validation/verify loop for the generated scaffold (Hard Rule 5 says it "must run" but no step instructs running the scaffold's tests to prove it), and no explicit handling of what to do if a user's answer in step 1 conflicts with later steps. Not a 3: the sequence is complete and coherent with most checkpoints present, and scaffold generation is not one of the destructive/batch contexts that would cap the score at 3.

4 / 5

Progressive Disclosure

Good structure for a single-file skill: the body is a well-organized overview (framework table, workflow, trade-offs, rules, four worked use cases) and the Upstream section clearly signals where the load-on-demand material lives — "5 load-on-demand reference docs and 3 runnable example scaffolds", the repo URL, and the npx install command. It stays below anchor 5 because those references are external (a GitHub URL) rather than one-level-deep local files — none of the reference docs, interfaces, or example scaffolds are actually present in the skill bundle, so a consumer must fetch the upstream repo to reach the detailed material. Not a 3: what is inlined (framework table, trade-off menus, use-case sketches) is appropriately overview-level, and navigation to details is clearly signaled.

4 / 5

Total

16

/

20

Passed

Description

83%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 well-constructed description with an explicit 'Use when' trigger clause, third-person voice, an enumerated six-stage framework, and a broad list of concrete triggering use cases. The only gaps are a single action verb rather than several distinct actions and a few missing natural synonyms (recommender system, personalized feed).

DimensionReasoningScore

Specificity

"Design composable recommendation, ranking, and feed pipelines using the six-stage Source→Hydrator→Filter→Scorer→Selector→SideEffect framework" names the domain and the concrete mechanism with all six stages enumerated, and the trigger list covers content feeds, search ranking, RAG rerankers, task prioritizers, notification triage, and ad selection. It stops short of anchor 5 because there is essentially one action verb ("Design") rather than multiple distinct concrete actions like the extract/fill/merge/convert exemplar; not a 3, since coverage of scope is comprehensive rather than partial.

4 / 5

Completeness

Both halves are explicit: the "what" is "Design composable recommendation, ranking, and feed pipelines using the six-stage... framework" and the "when" is a literal trigger clause — "Use when building any system that picks the top K items for a (user, context)" with six concrete triggering scenarios. This mirrors the anchor-5 exemplar structure (clear what + explicit 'Use when' with concrete trigger phrases); not a 4, because the 'when' is already explicit and specific rather than merely present.

5 / 5

Trigger Term Quality

Natural trigger phrases are strong: "content feeds, search ranking, RAG rerankers, task prioritizers, notification triage, ad selection" plus "top K items for a (user, context)" are phrases users would plausibly say. A few common variations are missing ("recommender system", "personalized feed", "For You feed", "candidate generation"), which keeps it below the comprehensive-synonym anchor 5 but comfortably above anchor 3's partial coverage.

4 / 5

Distinctiveness Conflict Risk

The six-stage pipeline framing and the "top K for (user, context)" scoping carve out a clear niche that would not fire for unrelated skills. Minor overlap risk remains with general RAG/search-ranking and task-management skills via the "RAG rerankers" and "task prioritizers" triggers, which keeps it below the minimal-conflict anchor 5 but above anchor 3's "could still overlap with similar skills".

4 / 5

Total

17

/

20

Passed

Validation

100%

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

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository

wshobson/agents

Path

plugins/machine-learning-ops/skills/recsys-pipeline-architect/SKILL.md

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.