skill-forge-recap

Name: skill-forge-recap
Author: WrathZA

17stars0forks

How to install skill-forge-recap?

wrathza/skillforge/skill-forge-recap

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

Installs into the current project.

Summary

Read a skill's body and report what it actually does vs. what the description claims: drift, undeclared behaviors, verdict. After the recap, offers context-sensitive actions: fix frontmatter via HITL, hand off to skill-forge-update, or run skill-forge-judge. Use before updating a

Files · 1

View on GitHub

SKILL.md

1	# Skill Forge Recap
2	<!-- Pattern: Tool (structured audit output) -->
3
4	Audit a skill by reading its actual implementation and reporting what it really does — independent of what the description claims. The description is a claim; the body is the ground truth.
5
6	---
7
8	## What to Read
9
10	Use the path provided or resolved by the environment. Environment-level path overrides (e.g. CLAUDE.md remapping) take precedence over any default path.
11
12	If the file doesn't exist, report: "Skill not found — check name spelling" and stop. Do not search for alternatives.
13
14	Also read any `references/` files marked MANDATORY READ — behaviors that live outside the main body still count.
15
16	If a `references/` directory exists but no files are marked MANDATORY READ, include this note in Undeclared behaviors: "references/ directory present but no MANDATORY READ triggers — possible Pattern 3 (Orphan References)."
17
18	---
19
20	## What to Compare
21
22	The description makes four implicit claims. Check each against the body:
23
24	1. WHAT it does — does the body do this, or something adjacent?
25	2. WHEN to use it — do trigger phrases match actual entry conditions?
26	3. Keywords — do searchable terms reflect what the body covers?
27	4. Scope — does the description under- or over-state coverage?
28
29	Also scan for undeclared behaviors: things the body does that the description never mentions. Focus on the locations where iterative edits silently accumulate new behavior: NEVER rules, phases/steps, skip/exit conditions, reference-loading triggers, and confirmation loops.
30
31	Before writing your verdict, ask: Would an agent picking skills from descriptions alone make the right call? If the description would misroute the skill, that's Significant drift regardless of how small the wording gap appears.
32
33	---
34
35	## Output Format
36
37	```
38	## Recap: <skill-name>
39
40	Description claims: [one-line summary of what description promises]
41
42	Body actually does: [one-line summary derived independently from reading the body]
43
44	Drift found:
45	- [specific discrepancy]
46	(or "None found")
47
48	Undeclared behaviors:
49	- [behavior in body not mentioned in description]
50	(or "None found")
51
52	Verdict: Aligned / Minor drift / Significant drift
53	```
54
55	---
56
57	## Post-Recap Actions
58
59	After writing the recap block, classify the findings into two buckets:
60
61	- Frontmatter drift: discrepancies in the `name` or `description` fields
62	- Body drift: behavioral discrepancies or undeclared behaviors
63
64	If neither bucket has findings, output "No drift found." and stop — do not show the action menu.
65
66	Otherwise, show only the options relevant to what was found:
67
68	```
69	(f)ix frontmatter — HITL through corrected name/description fields [only if frontmatter drift]
70	(u)pdate body — hand off to skill-forge-update with drift loaded [only if body drift]
71	(j)udge — run skill-forge-judge for quality scoring [always]
72	```
73
74	If both frontmatter and body drift are detected, show all applicable options together.
75
76	### (f) Fix frontmatter
77
78	Produce a numbered findings list covering only the drifted fields, formatted for skill-forge-hitl. Each item must include the current value and the corrected value:
79
80	```
81	1. name: "<current>" → "<corrected>"
82	2. description: "<current>" → "<corrected>"
83	```
84
85	Include only fields where drift was found. Then invoke `/skill-forge-hitl` on that list.
86
87	### (u) Update body
88
89	Invoke `/skill-forge-update` and pass the recap output as pre-loaded context. State explicitly that Phase 0 is already complete so skill-forge-update skips directly to Phase 1 change elicitation.
90
91	### (j) Judge
92
93	Invoke `/skill-forge-judge` on the recapped skill. This is always available — continuous refinement applies even when no drift was detected.
94
95	---
96
97	## Verdict Criteria
98
99	Minor drift: Description is imprecise but not misleading. The skill would still be selected for the right task.
100
101	Significant drift: Description would cause the skill to be selected for the wrong task, OR the agent would miss a key behavior because the description omits it.
102
103	An incomplete description (missing behaviors) differs from a wrong description (claiming behaviors that don't exist). Flag both, label them separately.
104
105	---
106
107	## NEVER
108
109	- NEVER start your "body actually does" summary by paraphrasing the description
110	Instead: Read the body first, form an independent summary, then compare to the description.
111	Why: Starting from the description anchors you to its framing — you'll miss drift because your summary will match it by construction.
112
113	- **NEVER skip MANDATORY READ ref

People who liked this also viewed

6 picks

Type

skill-forge-recap

17stars0forks

View on GitHub

How to install skill-forge-recap?

wrathza/skillforge/skill-forge-recap

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Skill Forge Recap
2	<!-- Pattern: Tool (structured audit output) -->
3
4	Audit a skill by reading its actual implementation and reporting what it really does — independent of what the description claims. The description is a claim; the body is the ground truth.
5
6	---
7
8	## What to Read
9
10	Use the path provided or resolved by the environment. Environment-level path overrides (e.g. CLAUDE.md remapping) take precedence over any default path.
11
12	If the file doesn't exist, report: "Skill not found — check name spelling" and stop. Do not search for alternatives.
13
14	Also read any `references/` files marked MANDATORY READ — behaviors that live outside the main body still count.
15
16	If a `references/` directory exists but no files are marked MANDATORY READ, include this note in Undeclared behaviors: "references/ directory present but no MANDATORY READ triggers — possible Pattern 3 (Orphan References)."
17
18	---
19
20	## What to Compare
21
22	The description makes four implicit claims. Check each against the body:
23
24	1. WHAT it does — does the body do this, or something adjacent?
25	2. WHEN to use it — do trigger phrases match actual entry conditions?
26	3. Keywords — do searchable terms reflect what the body covers?
27	4. Scope — does the description under- or over-state coverage?
28
29	Also scan for undeclared behaviors: things the body does that the description never mentions. Focus on the locations where iterative edits silently accumulate new behavior: NEVER rules, phases/steps, skip/exit conditions, reference-loading triggers, and confirmation loops.
30
31	Before writing your verdict, ask: Would an agent picking skills from descriptions alone make the right call? If the description would misroute the skill, that's Significant drift regardless of how small the wording gap appears.
32
33	---
34
35	## Output Format
36
37	```
38	## Recap: <skill-name>
39
40	Description claims: [one-line summary of what description promises]
41
42	Body actually does: [one-line summary derived independently from reading the body]
43
44	Drift found:
45	- [specific discrepancy]
46	(or "None found")
47
48	Undeclared behaviors:
49	- [behavior in body not mentioned in description]
50	(or "None found")
51
52	Verdict: Aligned / Minor drift / Significant drift
53	```
54
55	---
56
57	## Post-Recap Actions
58
59	After writing the recap block, classify the findings into two buckets:
60
61	- Frontmatter drift: discrepancies in the `name` or `description` fields
62	- Body drift: behavioral discrepancies or undeclared behaviors
63
64	If neither bucket has findings, output "No drift found." and stop — do not show the action menu.
65
66	Otherwise, show only the options relevant to what was found:
67
68	```
69	(f)ix frontmatter — HITL through corrected name/description fields [only if frontmatter drift]
70	(u)pdate body — hand off to skill-forge-update with drift loaded [only if body drift]
71	(j)udge — run skill-forge-judge for quality scoring [always]
72	```
73
74	If both frontmatter and body drift are detected, show all applicable options together.
75
76	### (f) Fix frontmatter
77
78	Produce a numbered findings list covering only the drifted fields, formatted for skill-forge-hitl. Each item must include the current value and the corrected value:
79
80	```
81	1. name: "<current>" → "<corrected>"
82	2. description: "<current>" → "<corrected>"
83	```
84
85	Include only fields where drift was found. Then invoke `/skill-forge-hitl` on that list.
86
87	### (u) Update body
88
89	Invoke `/skill-forge-update` and pass the recap output as pre-loaded context. State explicitly that Phase 0 is already complete so skill-forge-update skips directly to Phase 1 change elicitation.
90
91	### (j) Judge
92
93	Invoke `/skill-forge-judge` on the recapped skill. This is always available — continuous refinement applies even when no drift was detected.
94
95	---
96
97	## Verdict Criteria
98
99	Minor drift: Description is imprecise but not misleading. The skill would still be selected for the right task.
100
101	Significant drift: Description would cause the skill to be selected for the wrong task, OR the agent would miss a key behavior because the description omits it.
102
103	An incomplete description (missing behaviors) differs from a wrong description (claiming behaviors that don't exist). Flag both, label them separately.
104
105	---
106
107	## NEVER
108
109	- NEVER start your "body actually does" summary by paraphrasing the description
110	Instead: Read the body first, form an independent summary, then compare to the description.
111	Why: Starting from the description anchors you to its framing — you'll miss drift because your summary will match it by construction.
112
113	- **NEVER skip MANDATORY READ ref

People who liked this also viewed

6 picks

Type