agentic-engineering

Name: agentic-engineering
Author: affaan-m

232kstars35kforks

How to install agentic-engineering?

affaan-m/ecc/agentic-engineering

$npx -y skills add affaan-m/ECC --skill agentic-engineering

Installs into the current project.

Summary

Operate as an agentic engineer using eval-first execution, decomposition, and cost-aware model routing. Use when AI agents perform most implementation work and humans enforce quality and risk controls.

Files · 1

View on GitHub

SKILL.md

1	# Agentic Engineering
2
3	Use this skill for engineering workflows where AI agents perform most implementation work and humans enforce quality and risk controls.
4
5	## Operating Principles
6
7	1. Define completion criteria before execution.
8	2. Decompose work into agent-sized units.
9	3. Route model tiers by task complexity.
10	4. Measure with evals and regression checks.
11
12	## Eval-First Loop
13
14	1. Define capability eval and regression eval.
15	2. Run baseline and capture failure signatures.
16	3. Execute implementation.
17	4. Re-run evals and compare deltas.
18
19	Example workflow:
20	```
21	1. Write test that captures desired behavior (eval)
22	2. Run test → capture baseline failures
23	3. Implement feature
24	4. Re-run test → verify improvements
25	5. Check for regressions in other tests
26	```
27
28	## Task Decomposition
29
30	Apply the 15-minute unit rule:
31	- Each unit should be independently verifiable
32	- Each unit should have a single dominant risk
33	- Each unit should expose a clear done condition
34
35	Good decomposition:
36	```
37	Task: Add user authentication
38	├─ Unit 1: Add password hashing (15 min, security risk)
39	├─ Unit 2: Create login endpoint (15 min, API contract risk)
40	├─ Unit 3: Add session management (15 min, state risk)
41	└─ Unit 4: Protect routes with middleware (15 min, auth logic risk)
42	```
43
44	Bad decomposition:
45	```
46	Task: Add user authentication (2 hours, multiple risks)
47	```
48
49	## Model Routing
50
51	Choose model tier based on task complexity:
52
53	- Haiku: Classification, boilerplate transforms, narrow edits
54	- Example: Rename variable, add type annotation, format code
55
56	- Sonnet: Implementation and refactors
57	- Example: Implement feature, refactor module, write tests
58
59	- Opus: Architecture, root-cause analysis, multi-file invariants
60	- Example: Design system, debug complex issue, review architecture
61
62	Cost discipline: Escalate model tier only when lower tier fails with a clear reasoning gap.
63
64	## Session Strategy
65
66	- Continue session for closely-coupled units
67	- Example: Implementing related functions in same module
68
69	- Start fresh session after major phase transitions
70	- Example: Moving from implementation to testing
71
72	- Compact after milestone completion, not during active debugging
73	- Example: After feature complete, before starting next feature
74
75	## Review Focus for AI-Generated Code
76
77	Prioritize:
78	- Invariants and edge cases
79	- Error boundaries
80	- Security and auth assumptions
81	- Hidden coupling and rollout risk
82
83	Do not waste review cycles on style-only disagreements when automated format/lint already enforce style.
84
85	Review checklist:
86	- [ ] Edge cases handled (null, empty, boundary values)
87	- [ ] Error handling comprehensive
88	- [ ] Security assumptions validated
89	- [ ] No hidden coupling between modules
90	- [ ] Rollout risk assessed (breaking changes, migrations)
91
92	## Cost Discipline
93
94	Track per task:
95	- Model tier used
96	- Token estimate
97	- Retries needed
98	- Wall-clock time
99	- Success/failure outcome
100
101	Example tracking:
102	```
103	Task: Implement user login
104	Model: Sonnet
105	Tokens: ~5k input, ~2k output
106	Retries: 1 (initial implementation had auth bug)
107	Time: 8 minutes
108	Outcome: Success
109	```
110
111	## When to Use This Skill
112
113	- Managing AI-driven development workflows
114	- Planning agent task decomposition
115	- Optimizing model tier selection
116	- Implementing eval-first development
117	- Reviewing AI-generated code
118	- Tracking development costs
119
120	## Integration with Other Skills
121
122	- tdd-workflow: Combine with eval-first loop for test-driven development
123	- verification-loop: Use for continuous validation during implementation
124	- search-first: Apply before implementation to find existing solutions
125	- coding-standards: Reference during code review phase

People who liked this also viewed

6 picks

Type

agentic-engineering

232kstars35kforks

View on GitHub

How to install agentic-engineering?

affaan-m/ecc/agentic-engineering

$npx -y skills add affaan-m/ECC --skill agentic-engineering

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Agentic Engineering
2
3	Use this skill for engineering workflows where AI agents perform most implementation work and humans enforce quality and risk controls.
4
5	## Operating Principles
6
7	1. Define completion criteria before execution.
8	2. Decompose work into agent-sized units.
9	3. Route model tiers by task complexity.
10	4. Measure with evals and regression checks.
11
12	## Eval-First Loop
13
14	1. Define capability eval and regression eval.
15	2. Run baseline and capture failure signatures.
16	3. Execute implementation.
17	4. Re-run evals and compare deltas.
18
19	Example workflow:
20	```
21	1. Write test that captures desired behavior (eval)
22	2. Run test → capture baseline failures
23	3. Implement feature
24	4. Re-run test → verify improvements
25	5. Check for regressions in other tests
26	```
27
28	## Task Decomposition
29
30	Apply the 15-minute unit rule:
31	- Each unit should be independently verifiable
32	- Each unit should have a single dominant risk
33	- Each unit should expose a clear done condition
34
35	Good decomposition:
36	```
37	Task: Add user authentication
38	├─ Unit 1: Add password hashing (15 min, security risk)
39	├─ Unit 2: Create login endpoint (15 min, API contract risk)
40	├─ Unit 3: Add session management (15 min, state risk)
41	└─ Unit 4: Protect routes with middleware (15 min, auth logic risk)
42	```
43
44	Bad decomposition:
45	```
46	Task: Add user authentication (2 hours, multiple risks)
47	```
48
49	## Model Routing
50
51	Choose model tier based on task complexity:
52
53	- Haiku: Classification, boilerplate transforms, narrow edits
54	- Example: Rename variable, add type annotation, format code
55
56	- Sonnet: Implementation and refactors
57	- Example: Implement feature, refactor module, write tests
58
59	- Opus: Architecture, root-cause analysis, multi-file invariants
60	- Example: Design system, debug complex issue, review architecture
61
62	Cost discipline: Escalate model tier only when lower tier fails with a clear reasoning gap.
63
64	## Session Strategy
65
66	- Continue session for closely-coupled units
67	- Example: Implementing related functions in same module
68
69	- Start fresh session after major phase transitions
70	- Example: Moving from implementation to testing
71
72	- Compact after milestone completion, not during active debugging
73	- Example: After feature complete, before starting next feature
74
75	## Review Focus for AI-Generated Code
76
77	Prioritize:
78	- Invariants and edge cases
79	- Error boundaries
80	- Security and auth assumptions
81	- Hidden coupling and rollout risk
82
83	Do not waste review cycles on style-only disagreements when automated format/lint already enforce style.
84
85	Review checklist:
86	- [ ] Edge cases handled (null, empty, boundary values)
87	- [ ] Error handling comprehensive
88	- [ ] Security assumptions validated
89	- [ ] No hidden coupling between modules
90	- [ ] Rollout risk assessed (breaking changes, migrations)
91
92	## Cost Discipline
93
94	Track per task:
95	- Model tier used
96	- Token estimate
97	- Retries needed
98	- Wall-clock time
99	- Success/failure outcome
100
101	Example tracking:
102	```
103	Task: Implement user login
104	Model: Sonnet
105	Tokens: ~5k input, ~2k output
106	Retries: 1 (initial implementation had auth bug)
107	Time: 8 minutes
108	Outcome: Success
109	```
110
111	## When to Use This Skill
112
113	- Managing AI-driven development workflows
114	- Planning agent task decomposition
115	- Optimizing model tier selection
116	- Implementing eval-first development
117	- Reviewing AI-generated code
118	- Tracking development costs
119
120	## Integration with Other Skills
121
122	- tdd-workflow: Combine with eval-first loop for test-driven development
123	- verification-loop: Use for continuous validation during implementation
124	- search-first: Apply before implementation to find existing solutions
125	- coding-standards: Reference during code review phase

People who liked this also viewed

6 picks

Type