@@ -24,20 +24,26 @@ This isn't because the model isn't smart enough. It's because the working enviro
2424
2525``` mermaid
2626flowchart TB
27- Spec["Task not specific<br/>'add search' can mean many things"] --> Fail["Result:<br/>agent writes code<br/>but work is unreliable"]
28- Context["Project rules missing<br/>agent cannot see them"] --> Fail
29- Env["Environment broken<br/>deps or commands fail"] --> Fail
30- Verify["No tests or check commands"] --> Fail
31- State["No progress file<br/>next session starts blind"] --> Fail
32- Fail --> Fix["Fix the broken layer,<br/>then rerun"]
27+ Start["Start the task"] --> Q1{"Is the task specific?"}
28+ Q1 -->|No| Bad["Result:<br/>lots of code<br/>still unstable"]
29+ Q1 -->|Yes| Q2{"Are the rules in the repo?"}
30+ Q2 -->|No| Bad
31+ Q2 -->|Yes| Q3{"Does the environment run?"}
32+ Q3 -->|No| Bad
33+ Q3 -->|Yes| Q4{"Are there check commands?"}
34+ Q4 -->|No| Bad
35+ Q4 -->|Yes| Q5{"Is progress recorded?"}
36+ Q5 -->|No| Bad
37+ Q5 -->|Yes| Good["More likely to finish cleanly"]
3338```
3439
3540``` mermaid
36- flowchart LR
37- Same["Same prompt<br/>same model"] --> Bare["Bare run<br/>20 min / $9<br/>core features broken"]
38- Same --> Full["Add planning, checks, and evaluation<br/>6 hr / $200<br/>playable app"]
39- Bare --> Lesson["Only the harness changed"]
40- Full --> Lesson
41+ flowchart TB
42+ Same["Same task<br/>same model"]
43+ Same --> Bare["Just one prompt<br/>20 min / $9"]
44+ Same --> Full["Rules and checks first<br/>6 hr / $200"]
45+ Bare --> Bad["Core features broken"]
46+ Full --> Good["Playable app"]
4147```
4248
4349## Why This Happens
0 commit comments