You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: .github/ISSUE_TEMPLATE/propose-anchor.yml
+26-5Lines changed: 26 additions & 5 deletions
Original file line number
Diff line number
Diff line change
@@ -11,12 +11,33 @@ body:
11
11
Thank you for proposing a new semantic anchor! Here's the workflow:
12
12
13
13
1. **You provide a term or concept** that you believe would make a good semantic anchor
14
-
2. **GitHub Copilot validates** whether it meets the quality criteria (precise, rich, consistent, attributable)
15
-
3. **If accepted**: Copilot enriches this issue with detailed information (proponents, categories, related concepts)
16
-
4. **If rejected**: Copilot explains why it doesn't meet the criteria
17
-
5. **After enrichment**: Copilot is assigned to create the anchor file and submit a PR
14
+
2. **The review agent validates** whether it meets the quality criteria (see below)
15
+
3. **If accepted**: The agent enriches this issue with detailed information (proponents, categories, related concepts) and assigns a quality tier
16
+
4. **If rejected**: The agent explains why it doesn't meet the criteria
17
+
5. **After enrichment**: The agent creates the anchor file and submits a PR
18
18
19
-
**Just provide the term below - Copilot handles the rest!**
19
+
### Quality Criteria
20
+
21
+
A semantic anchor must be **precise**, **rich**, **consistent**, and **attributable**.
22
+
23
+
### Tier Rating (★)
24
+
25
+
The review agent assigns one of three tiers based on how reliably the anchor improves LLM communication:
26
+
27
+
- **★★★ Self-standing**: Saying the anchor name alone reliably activates the correct behavior in the LLM. Clear, actionable, deterministic output. Examples: arc42, SOLID, Gherkin, MECE, Chain of Thought.
28
+
- **★★☆ Needs qualification**: The concept is known but requires a qualifier for reliable results. The anchor documentation must specify the recommended qualified form. Examples: "ADR according to Nygard", "3-point Pugh Matrix".
29
+
- **★☆☆ Descriptive only**: The term describes a concept but does not give the LLM clear instructions. These are **rejected** unless the proposer can demonstrate a concrete, actionable prompt pattern.
30
+
31
+
### Rejection Criteria
32
+
33
+
The review agent **rejects** proposals that are:
34
+
- Pure philosophy without actionable instructions (e.g., "TIMTOWTDI")
35
+
- Personality/assessment frameworks without LLM utility (e.g., "MBTI")
36
+
- Fully covered by an existing anchor (e.g., "Rubber Duck Debugging" ≈ Socratic Method)
37
+
- Too vague to produce consistent LLM output (e.g., "Keep it simple")
38
+
- Not an established concept (invented by the proposer)
39
+
40
+
**Just provide the term below - the review agent handles the rest!**
0 commit comments