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
- **Self-validating**: The validation script ensures all tests work correctly
457
-
- **Comprehensive**: Covers all aspects of functionality and interface design
458
-
- **Maintainable**: Clear structure and documentation for future updates
459
-
- **Scalable**: Tests can be added incrementally as functionality is implemented
460
-
- **Security-focused**: Security regression tests prevent reintroduction of vulnerabilities
461
-
462
434
### Visual Regression Testing
463
435
464
436
Visual regression tests ensure console output formatting remains consistent across code changes. The system uses golden files to capture expected output for table layouts, box rendering, tree structures, and error formatting.
@@ -581,14 +553,6 @@ graph TD
581
553
| **Workflow Health Manager** | `workflow-health-manager.md` | Monitor workflow health | Daily |
Copy file name to clipboardExpand all lines: .github/aw/agentic-chat.md
+2-24Lines changed: 2 additions & 24 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -32,10 +32,8 @@ Before assisting users, load and understand these instruction files from the gh-
32
32
- Let the coding agent determine implementation details
33
33
34
34
### 3. Problem Decomposition
35
-
Break down tasks into clear, actionable steps:
36
35
37
-
#### Step Structure
38
-
Provide clear, actionable steps that include:
36
+
Steps must include:
39
37
- What needs to be done
40
38
- Expected inputs and outputs
41
39
- Constraints or considerations
@@ -68,8 +66,6 @@ When creating task descriptions, follow this structure:
68
66
69
67
## Pseudo-Code Guidelines
70
68
71
-
When pseudo-code is necessary to clarify logic:
72
-
73
69
**Allowed**:
74
70
```
75
71
IF condition THEN
@@ -98,33 +94,15 @@ When you provide the final task description for the user to use, wrap it in **5
98
94
99
95
**Important**: The task title must start with "create a github agentic workflow that:" to trigger loading the appropriate instructions.
100
96
101
-
This allows users to:
102
-
1. Select the entire content between the 5-backtick blocks
103
-
2. Copy it directly
104
-
3. Paste it into a GitHub issue, pull request, or workflow file
105
-
106
97
## Interaction Guidelines
107
98
108
-
1.**Clarify Requirements**: Ask questions to understand the user's needs before generating a task description
99
+
1.**Clarify Requirements**: Ask about expected outcome, available context (repository, issue numbers), constraints, and tools needed (GitHub API, web search, file editing, etc.)
109
100
2.**Validate Understanding**: Summarize what you understand before creating the specification
110
101
3.**Iterate**: Be prepared to refine the task description based on user feedback
111
102
4.**Stay Focused**: Keep discussions centered on task specification, not implementation
112
103
5.**Reference Documentation**: Cite the loaded instruction files when relevant
113
104
6.**Summarize Updates**: On each chat turn after the initial request, provide a brief summary of the updates or changes provided by the user in the previous message, rather than re-reading the entire markdown content unless explicitly requested
114
105
115
-
## Example Interaction Flow
116
-
117
-
1. User describes a problem or task
118
-
2. You ask clarifying questions about:
119
-
- Expected outcome
120
-
- Available context (repository, issue numbers, etc.)
121
-
- Constraints or requirements
122
-
- Tools needed (GitHub API, web search, file editing, etc.)
123
-
3. You summarize your understanding
124
-
4. You generate a structured task description
125
-
5. You present it wrapped in 5 backticks for easy copy/paste
126
-
6. On subsequent turns, begin by summarizing the user's latest updates before making changes
127
-
128
106
## Terminology
129
107
130
108
Use correct terminology from the gh-aw project (see dictation instructions):
description: Guide for leveraging skills (SKILL.md files) in agentic workflows — hint strategy and fusion strategy
3
+
---
4
+
5
+
# Skills in Agentic Workflows
6
+
7
+
Consult this file when you want a workflow to take advantage of skills — domain-specific knowledge files (`SKILL.md`) that live in the repository under `skills/` or `.github/skills/`.
List available skills and their locations before deciding which strategy to apply.
20
+
21
+
---
22
+
23
+
## Strategy 1 — Hint (Generalist)
24
+
25
+
**Use when**: The task strategy is not fully known at authoring time, or when the agent must adapt to whatever skills are available.
26
+
27
+
The workflow prompt hints that skills exist and asks the agent to discover and apply the relevant ones itself. The agent decides which skill files to read and how much of each to use.
28
+
29
+
**Pattern**:
30
+
31
+
```markdown
32
+
If the repository contains `SKILL.md` files under `skills/`, check which ones are
33
+
relevant to this task. For each relevant skill, read its content and apply the
34
+
guidance it provides.
35
+
```
36
+
37
+
**When to prefer this**:
38
+
- The task domain is broad or open-ended
39
+
- You don't know at authoring time which skills will exist
40
+
- You want the agent to self-select relevant skills from a growing library
41
+
- The overhead of reading an extra file is acceptable
42
+
43
+
**Tradeoff**: The agent may read and summarise more skill content than needed, which costs context tokens and can dilute focus.
44
+
45
+
---
46
+
47
+
## Strategy 2 — Fusion (Ultra-Cognitive)
48
+
49
+
**Use when**: You know exactly which skill (or which part of a skill) is needed, and you want to minimise context overhead.
50
+
51
+
Extract and inline **only the specific sections** of the skill content that the agent needs. Do not paste the entire SKILL.md; identify the minimal fragment, then remix it into the workflow prompt so the agent receives precise, surgical guidance without loading the full file.
|**Scale**| Poor (entire skills loaded) | Good (minimal content) |
82
+
83
+
Fusion scales better because entire skills are never loaded. Prefer fusion when you know the task domain and the specific skill sections required.
84
+
85
+
---
86
+
87
+
## Example: Hint Strategy
88
+
89
+
```markdown
90
+
---
91
+
on:
92
+
issues:
93
+
types: [opened]
94
+
engine: copilot
95
+
tools:
96
+
github:
97
+
toolsets: [issues]
98
+
permissions:
99
+
issues: write
100
+
---
101
+
102
+
Triage the newly opened issue.
103
+
104
+
If there are relevant skills under `skills/`, read them and apply their guidance.
105
+
Focus on skills related to issue classification or project conventions.
106
+
```
107
+
108
+
---
109
+
110
+
## Example: Fusion Strategy
111
+
112
+
```markdown
113
+
---
114
+
on:
115
+
pull_request:
116
+
types: [opened, synchronize]
117
+
engine: copilot
118
+
tools:
119
+
github:
120
+
toolsets: [pull_requests]
121
+
permissions:
122
+
pull-requests: write
123
+
---
124
+
125
+
Review the pull request for adherence to project conventions.
126
+
127
+
<!-- Fused from skills/developer/SKILL.md#code-organization -->
128
+
Prefer many smaller files grouped by functionality. Add new files for new features
129
+
rather than extending existing ones. Keep validators under 300 lines; split when
130
+
a single file covers more than one domain.
131
+
<!-- End fusion -->
132
+
133
+
Report findings as inline review comments.
134
+
```
135
+
136
+
---
137
+
138
+
## Anti-Patterns
139
+
140
+
- ❌ **Do not load entire skill files** when only one section is relevant — use fusion instead
141
+
- ❌ **Do not hint without bounds** — if using the hint strategy, constrain the agent with a `maxdepth` and a relevance filter to avoid reading every SKILL.md in a large repo
142
+
- ❌ **Do not paste skills verbatim** without adapting them to the workflow's context — fused fragments should read as natural prose, not as lifted documentation
143
+
- ❌ **Do not hard-code skill file paths** in hints — use `find` so the prompt still works when skills are reorganised
0 commit comments