guidance

Control LLM output with regex and grammars, guarantee valid JSON/XML/code generation, enforce structured formats, and build multi-step workflows with Guidance - Microsoft Research's constrained generation framework

Quality

66%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./16-prompt-engineering/guidance/SKILL.md

Quality

Content

65%

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

The body is highly actionable with extensive executable examples, but it is verbose, lacks validation checkpoints in its multi-step patterns, and keeps far too much detail inline instead of pushing it to the existing reference files.

Suggestions

Trim conceptual explanations Claude already knows (token-healing problem/solution, "How it works" bullets, the unverified "30-50% faster" performance claims) and remove repeated model-loading boilerplate from every example.

Add explicit validate→fix→retry checkpoints to the ReAct and multi-step reasoning patterns, or note that these are generation recipes without destructive steps.

Move backend configuration, the full pattern catalog, and the comparison table into the existing references and signal them inline (e.g., "See references/backends.md for backend setup") so SKILL.md stays a lean overview.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient executable code but is padded with repeated model-loading boilerplate, conceptual explanations Claude already knows (the token-healing problem/solution, "How it works" bullets), and unverified performance claims; it could be tightened considerably, so not a 3, and not a 1 because the bulk is genuinely actionable code rather than prose padding.	2 / 3
Actionability	Provides numerous concrete, copy-paste-ready Python examples with real imports and specific calls (gen, select, context managers, @guidance functions) across many patterns, matching the fully-executable anchor.	3 / 3
Workflow Clarity	Multi-step patterns (ReAct agent, chain-of-thought, data extraction) show code-level sequencing but lack explicit validation checkpoints or validate-fix-retry feedback loops, which caps workflow clarity at 2; not a 1 because steps are clearly ordered within each function.	2 / 3
Progressive Disclosure	Bundle references (constraints.md, backends.md, examples.md) exist and are listed in a "See Also" section, but the body is a monolithic ~560-line wall that keeps backend config, all patterns, comparisons, and best practices inline — content that should be split out and signaled inline; not a 3 because references are not clearly signaled at the relevant sections and the overview is not lean.	2 / 3
	Total	9 / 12 Passed

Description

67%

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

The description is specific and names a distinct framework niche, but it omits any explicit "Use when..." trigger guidance and leans on technical jargon, which limits completeness and trigger-term quality.

Suggestions

Add an explicit trigger clause, e.g. "Use when you need to guarantee valid JSON/XML/code output, enforce regex or grammar constraints, or build multi-step constrained generation workflows."

Soften jargon ("grammars", "constrained generation framework") with more natural user-facing terms like "structured output" and "format enforcement" to broaden trigger coverage.

Include common variations users actually say (e.g. "structured output", "force valid JSON", "regex-constrained generation") to raise trigger-term quality.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — "Control LLM output with regex and grammars", "guarantee valid JSON/XML/code generation", "enforce structured formats", "build multi-step workflows" — matching the multiple-specific-actions anchor; not below because it goes well beyond naming a domain.	3 / 3
Completeness	Strongly answers "what" but contains no "Use when..." clause or equivalent explicit trigger guidance, so per the rubric guideline completeness is capped at 2; not a 1 because the what-side is comprehensive.	2 / 3
Trigger Term Quality	Covers some natural terms users would say ("valid JSON/XML", "regex", "structured formats") but leans technical with jargon like "grammars" and "constrained generation framework", missing common user variations; not a 3 because the keyword set is not broad and natural enough.	2 / 3
Distinctiveness Conflict Risk	Names a specific framework ("Guidance - Microsoft Research's constrained generation framework") with a distinct grammar/regex-constraint niche unlikely to trigger the wrong skill; not a 2 because the framework name and niche clearly separate it from generic structured-output skills.	3 / 3
	Total	10 / 12 Passed

Validation

87%

Warnings & errors only

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

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (573 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: Orchestra-Research/AI-Research-SKILLs
Path: 16-prompt-engineering/guidance/SKILL.md
Commit: 773a529

Reviewed: about 12 hours ago

Table of Contents

Discovery Implementation Validation

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.