skill-forge-judge

Name: skill-forge-judge
Author: WrathZA

17stars0forks

How to install skill-forge-judge?

wrathza/skillforge/skill-forge-judge

$npx -y skills add WrathZA/skillforge --skill skill-forge-judge

Installs into the current project.

Summary

Evaluate any LLM prompt (SKILL.md, CLAUDE.md, system prompts, bash guidance) for quality: grouped dimensional scoring with letter grade and step-through-ready numbered improvements list. Triggers: judge/review/audit/score/evaluate this skill or prompt.

Files · 1

View on GitHub

SKILL.md

1	# Skill Forge Judge
2
3	Evaluate any LLM-consumed prompt against quality standards, focused on knowledge delta, instruction clarity, and practical usability.
4
5	---
6
7	## Core Philosophy
8
9	> Good Prompt = Expert-only Knowledge − What Claude Already Knows
10
11	Restating defaults is token waste.
12
13	### Three Types of Knowledge
14
15	\| Type \| Definition \| Treatment \|
16	\|------\|------------\|-----------\|
17	\| Expert \| Claude genuinely doesn't know this \| Must keep — this is the value \|
18	\| Activation \| Claude knows but may not think of \| Keep if brief — serves as reminder \|
19	\| Redundant \| Claude definitely knows this \| Delete — wastes tokens \|
20
21	Good prompt: >70% Expert, <20% Activation, <10% Redundant.
22
23
24	## Evaluation Dimensions
25
26	Dimensions are grouped. Universal dimensions always apply. Type-specific modules apply based on what the prompt is. Multiple groups can apply to a single prompt.
27
28	Final grade = total score / total applicable points (as %)
29
30	\| Grade \| % \| Meaning \|
31	\|-------\|---\|---------\|
32	\| A \| 90%+ \| Excellent — production-ready \|
33	\| B \| 80–89% \| Good — minor improvements needed \|
34	\| C \| 70–79% \| Adequate — clear improvement path \|
35	\| D \| 60–69% \| Below average — significant issues \|
36	\| F \| <60% \| Poor — needs fundamental redesign \|
37
38	### Group U: Universal (80 pts) — always scored
39
40	MANDATORY — READ [`references/universal-dimensions.md`](references/universal-dimensions.md)
41
42	\| ID \| Dimension \| Pts \|
43	\|----\|-----------\|-----\|
44	\| U1 \| Knowledge/Instruction Delta \| 20 \|
45	\| U2 \| Mindset + Procedures \| 15 \|
46	\| U3 \| Anti-Pattern Quality \| 15 \|
47	\| U4 \| Freedom Calibration \| 15 \|
48	\| U5 \| Practical Usability \| 15 \|
49
50	### Group S: Skill Module (40 pts) — SKILL.md targets only
51
52	MANDATORY — READ [`references/skill-dimensions.md`](references/skill-dimensions.md)
53
54	\| ID \| Dimension \| Pts \|
55	\|----\|-----------\|-----\|
56	\| S1 \| Specification Compliance \| 15 \|
57	\| S2 \| Progressive Disclosure \| 15 \|
58	\| S3 \| Pattern Recognition \| 10 \|
59
60	### Group C: CLAUDE.md / System Prompt Module (40 pts) — prompts that govern LLM behavior settings
61
62	MANDATORY — READ [`references/claude-md-dimensions.md`](references/claude-md-dimensions.md)
63
64	\| ID \| Dimension \| Pts \|
65	\|----\|-----------\|-----\|
66	\| C1 \| Behavioral Clarity \| 15 \|
67	\| C2 \| Scope Definition \| 15 \|
68	\| C3 \| Structural Organization \| 10 \|
69
70	### Group B: Bash/Shell Module (30 pts) — prompts that contain shell/CLI guidance
71
72	MANDATORY — READ [`references/bash-dimensions.md`](references/bash-dimensions.md)
73
74	\| ID \| Dimension \| Pts \|
75	\|----\|-----------\|-----\|
76	\| B1 \| Rule Specificity & WHY \| 10 \|
77	\| B2 \| Anti-Pattern Coverage \| 10 \|
78	\| B3 \| Scope & Exceptions \| 10 \|
79
80	---
81
82	## Evaluation Protocol
83
84	> Evaluator's Lens: Before reading the target, adopt this question: "Does Claude already know this?" — every section gets marked [E], [A], or [R] before scoring begins.
85
86	### Step 0: Detect Prompt Type
87
88	Read the target and identify which groups apply:
89
90	```
91	[ ] Is it a SKILL.md file? → Score U + S. Do NOT load claude-md-dimensions.md or bash-dimensions.md.
92	MANDATORY: Load agentskills spec via Spec Cache (below) before scoring S1.
93	[ ] Is it a CLAUDE.md / system prompt? → Score U + C. Do NOT load skill-dimensions.md.
94	[ ] Does it contain bash/shell rules? → Also score B. Load bash-dimensions.md.
95	[ ] Is it something else? → Score U only. Do NOT load any type-specific reference.
96	```
97
98	Multiple groups can apply (e.g. a CLAUDE.md with bash guidance → U + C + B).
99
100	Edge cases:
101	- SKILL.md that also contains bash guidance → U + S + B
102	- A referenced sub-file (e.g. `references/bash.md`, not a root prompt) → U only; note "sub-file, not root prompt" in report
103	- Ambiguous type (could be CLAUDE.md or skill) → score both C and S groups; note the ambiguity in the Summary
104
105	### Spec Cache
106
107	The agentskills.io specification is cached daily so evaluations don't pay a network fetch on every run.
108
109	```
110	Cache dir: ~/.claude/tmp/
111	Filename: agentskills-spec-YYYY-MM-DD.md (today's date, e.g. agentskills-spec-2026-04-14.md)
112
113	1. Glob ~/.claude/tmp/agentskills-spec-<today>.md
114	2. If found → Read it — done.
115	3. If not → Bash: mkdir -p ~/.claude/tmp
116	WebFetch https://agentskills.io/specification
117	If WebFetch fails → score S1 from training knowledge; note "spec unavailable" in S1 Notes cell — done.
118	Write result to ~/.claude/tmp/agentskills-spec-<today>.md
119	Read the file — done.
120	```
121
122	Only fetch when the target is a SKILL.md (S1 scoring). Skip entirely for CLAUDE.md / system prompt / other evaluations.
123
124	### Step 1: First Pass — Knowledge Delta Scan
125
126	Read completely. Mark each section [E] Expert

People who liked this also viewed

6 picks

Type

skill-forge-judge

17stars0forks

View on GitHub

How to install skill-forge-judge?

wrathza/skillforge/skill-forge-judge

$npx -y skills add WrathZA/skillforge --skill skill-forge-judge

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Skill Forge Judge
2
3	Evaluate any LLM-consumed prompt against quality standards, focused on knowledge delta, instruction clarity, and practical usability.
4
5	---
6
7	## Core Philosophy
8
9	> Good Prompt = Expert-only Knowledge − What Claude Already Knows
10
11	Restating defaults is token waste.
12
13	### Three Types of Knowledge
14
15	\| Type \| Definition \| Treatment \|
16	\|------\|------------\|-----------\|
17	\| Expert \| Claude genuinely doesn't know this \| Must keep — this is the value \|
18	\| Activation \| Claude knows but may not think of \| Keep if brief — serves as reminder \|
19	\| Redundant \| Claude definitely knows this \| Delete — wastes tokens \|
20
21	Good prompt: >70% Expert, <20% Activation, <10% Redundant.
22
23
24	## Evaluation Dimensions
25
26	Dimensions are grouped. Universal dimensions always apply. Type-specific modules apply based on what the prompt is. Multiple groups can apply to a single prompt.
27
28	Final grade = total score / total applicable points (as %)
29
30	\| Grade \| % \| Meaning \|
31	\|-------\|---\|---------\|
32	\| A \| 90%+ \| Excellent — production-ready \|
33	\| B \| 80–89% \| Good — minor improvements needed \|
34	\| C \| 70–79% \| Adequate — clear improvement path \|
35	\| D \| 60–69% \| Below average — significant issues \|
36	\| F \| <60% \| Poor — needs fundamental redesign \|
37
38	### Group U: Universal (80 pts) — always scored
39
40	MANDATORY — READ [`references/universal-dimensions.md`](references/universal-dimensions.md)
41
42	\| ID \| Dimension \| Pts \|
43	\|----\|-----------\|-----\|
44	\| U1 \| Knowledge/Instruction Delta \| 20 \|
45	\| U2 \| Mindset + Procedures \| 15 \|
46	\| U3 \| Anti-Pattern Quality \| 15 \|
47	\| U4 \| Freedom Calibration \| 15 \|
48	\| U5 \| Practical Usability \| 15 \|
49
50	### Group S: Skill Module (40 pts) — SKILL.md targets only
51
52	MANDATORY — READ [`references/skill-dimensions.md`](references/skill-dimensions.md)
53
54	\| ID \| Dimension \| Pts \|
55	\|----\|-----------\|-----\|
56	\| S1 \| Specification Compliance \| 15 \|
57	\| S2 \| Progressive Disclosure \| 15 \|
58	\| S3 \| Pattern Recognition \| 10 \|
59
60	### Group C: CLAUDE.md / System Prompt Module (40 pts) — prompts that govern LLM behavior settings
61
62	MANDATORY — READ [`references/claude-md-dimensions.md`](references/claude-md-dimensions.md)
63
64	\| ID \| Dimension \| Pts \|
65	\|----\|-----------\|-----\|
66	\| C1 \| Behavioral Clarity \| 15 \|
67	\| C2 \| Scope Definition \| 15 \|
68	\| C3 \| Structural Organization \| 10 \|
69
70	### Group B: Bash/Shell Module (30 pts) — prompts that contain shell/CLI guidance
71
72	MANDATORY — READ [`references/bash-dimensions.md`](references/bash-dimensions.md)
73
74	\| ID \| Dimension \| Pts \|
75	\|----\|-----------\|-----\|
76	\| B1 \| Rule Specificity & WHY \| 10 \|
77	\| B2 \| Anti-Pattern Coverage \| 10 \|
78	\| B3 \| Scope & Exceptions \| 10 \|
79
80	---
81
82	## Evaluation Protocol
83
84	> Evaluator's Lens: Before reading the target, adopt this question: "Does Claude already know this?" — every section gets marked [E], [A], or [R] before scoring begins.
85
86	### Step 0: Detect Prompt Type
87
88	Read the target and identify which groups apply:
89
90	```
91	[ ] Is it a SKILL.md file? → Score U + S. Do NOT load claude-md-dimensions.md or bash-dimensions.md.
92	MANDATORY: Load agentskills spec via Spec Cache (below) before scoring S1.
93	[ ] Is it a CLAUDE.md / system prompt? → Score U + C. Do NOT load skill-dimensions.md.
94	[ ] Does it contain bash/shell rules? → Also score B. Load bash-dimensions.md.
95	[ ] Is it something else? → Score U only. Do NOT load any type-specific reference.
96	```
97
98	Multiple groups can apply (e.g. a CLAUDE.md with bash guidance → U + C + B).
99
100	Edge cases:
101	- SKILL.md that also contains bash guidance → U + S + B
102	- A referenced sub-file (e.g. `references/bash.md`, not a root prompt) → U only; note "sub-file, not root prompt" in report
103	- Ambiguous type (could be CLAUDE.md or skill) → score both C and S groups; note the ambiguity in the Summary
104
105	### Spec Cache
106
107	The agentskills.io specification is cached daily so evaluations don't pay a network fetch on every run.
108
109	```
110	Cache dir: ~/.claude/tmp/
111	Filename: agentskills-spec-YYYY-MM-DD.md (today's date, e.g. agentskills-spec-2026-04-14.md)
112
113	1. Glob ~/.claude/tmp/agentskills-spec-<today>.md
114	2. If found → Read it — done.
115	3. If not → Bash: mkdir -p ~/.claude/tmp
116	WebFetch https://agentskills.io/specification
117	If WebFetch fails → score S1 from training knowledge; note "spec unavailable" in S1 Notes cell — done.
118	Write result to ~/.claude/tmp/agentskills-spec-<today>.md
119	Read the file — done.
120	```
121
122	Only fetch when the target is a SKILL.md (S1 scoring). Skip entirely for CLAUDE.md / system prompt / other evaluations.
123
124	### Step 1: First Pass — Knowledge Delta Scan
125
126	Read completely. Mark each section [E] Expert

People who liked this also viewed

6 picks

Type