fix: strengthen code generation discipline in system prompt and write_file tool

System prompt (main.go):
- Added 'Code generation discipline' section with 6 rules:
  - Use EXACT file paths from the task (never simplify)
  - Never modify source files when creating new output files
  - Write correct code on first attempt (no write-test-rewrite loops)
  - Follow specified function signatures and designs exactly
  - After writing a file, do not run tests unless asked

write_file tool (file_tool.go):
- Added CRITICAL note to description about exact path usage
- Updated schema path parameter to warn against path simplification

Author: molty3000

cmd/odek/file_tool.go

@@ -157,7 +157,8 @@ func (t *writeFileTool) Name() string { return "write_file" }
 func (t *writeFileTool) Description() string {
 	return `Write content to a file, completely replacing existing content.
 Creates parent directories automatically. OVERWRITES the entire file.
-Use patch for targeted edits.`
+Use patch for targeted edits.
+CRITICAL: Use the EXACT path specified in the task. Do not simplify or drop directories from the path.`
 }
 
 func (t *writeFileTool) Schema() any {
@@ -166,7 +167,7 @@ func (t *writeFileTool) Schema() any {
 		"properties": map[string]any{
 			"path": map[string]any{
 				"type":        "string",
-				"description": "Path to the file to write (will be created/overwritten).",
+				"description": "Path to the file to write (will be created/overwritten). Use the EXACT path from the task — never drop or simplify directories.",
 			},
 			"content": map[string]any{
 				"type":        "string",

cmd/odek/main.go

@@ -97,6 +97,20 @@ Output discipline:
 - When counting or measuring (LOC, files, etc.), double-check your numbers.
   Run the counting command and verify the output before answering.
 
+Code generation discipline — critical for writing files:
+- When the task specifies an output file path (e.g. "benchmark_data/output/X.py"),
+  use that EXACT path with write_file. Never drop directories or simplify.
+  A path like "benchmark_data/output/X.py" must be written as "benchmark_data/output/X.py",
+  not "X.py" or "output/X.py".
+- NEVER modify existing source files when the task asks you to CREATE new output files.
+  Read-only operations on source files are fine; just don't write to them.
+- Write correct code on the FIRST attempt. Do NOT write, then test, then rewrite, then retest.
+  If you need to verify, read the file back once, then move on.
+- Follow the specified function signature, class structure, and design EXACTLY.
+  Do not invent a different architecture or framework than what was asked for.
+- After writing a file, do NOT run tests or verify with shell commands unless
+  the task explicitly asks you to. One write is enough.
+
 Tool conventions — use these dedicated tools, NOT shell commands:
 - Do NOT use cat/head/tail to read files — use read_file instead (line numbers, pagination).
 - Do NOT use grep/rg/find to search — use search_files instead (regex, glob, context lines).