gpt-engineer-app[bot] commited on
Commit
01de83c
Β·
1 Parent(s): f9a3fa6
src/server/branch-debug.functions.ts CHANGED
@@ -231,7 +231,7 @@ const analysisTool = {
231
  },
232
  };
233
 
234
- const SYSTEM_PROMPT = `You are BranchDebug, a code-aware root-cause analyzer. Inputs are:
235
  1. A SANITIZED unified git diff where real identifiers have been replaced with opaque tokens like fn_0019.
236
  2. A natural-language description of an observed failure.
237
  3. A numbered list of HUNKS (filePath, lineRange, function context).
@@ -367,7 +367,7 @@ const snippetTool = {
367
  },
368
  };
369
 
370
- const SNIPPET_SYSTEM = `You are BranchDebug in SNIPPET mode β€” an expert code reviewer with deep knowledge of every mainstream programming language (Python, TypeScript/JavaScript, C/C++, C#, Java, Kotlin, Swift, Go, Rust, Ruby, PHP, Scala, Elixir, Haskell, Lua, R, Dart, SQL, Bash, HTML/CSS, YAML/JSON/TOML, and more).
371
 
372
  The user pasted a raw code snippet (not a diff). Identifiers are tokenized as fn_NNNN; treat them as opaque names. Carefully analyze the snippet and find ANY of the following classes of bugs that match the failure description (or are obvious defects, even if not described):
373
 
 
231
  },
232
  };
233
 
234
+ const SYSTEM_PROMPT = `You are BranchDebug Bot, a code-aware root-cause analyzer powered by Qwen3 reasoning. Inputs are:
235
  1. A SANITIZED unified git diff where real identifiers have been replaced with opaque tokens like fn_0019.
236
  2. A natural-language description of an observed failure.
237
  3. A numbered list of HUNKS (filePath, lineRange, function context).
 
367
  },
368
  };
369
 
370
+ const SNIPPET_SYSTEM = `You are BranchDebug Bot in SNIPPET mode, powered by Qwen3 reasoning β€” an expert code reviewer with deep knowledge of every mainstream programming language (Python, TypeScript/JavaScript, C/C++, C#, Java, Kotlin, Swift, Go, Rust, Ruby, PHP, Scala, Elixir, Haskell, Lua, R, Dart, SQL, Bash, HTML/CSS, YAML/JSON/TOML, and more).
371
 
372
  The user pasted a raw code snippet (not a diff). Identifiers are tokenized as fn_NNNN; treat them as opaque names. Carefully analyze the snippet and find ANY of the following classes of bugs that match the failure description (or are obvious defects, even if not described):
373
 
src/server/forensic.functions.ts CHANGED
@@ -289,14 +289,14 @@ export const fetchVehicleCode = createServerFn({ method: "POST" })
289
  // ───────────────────────── Stage analysis ─────────────────────────
290
 
291
  const STAGE_PROMPTS: Record<1 | 2 | 3, string> = {
292
- 1: `You are an expert AV engineer specializing in post-deployment forensic debugging.
293
  Inputs: deployed vehicle code (identifiers anonymized as fn_NNNN) and a failure description.
294
  Logs are NOT yet available. Analyze the code statically. Generate ranked hypotheses and predict log signatures.
295
  Always call submit_stage1.`,
296
- 2: `You are an expert AV engineer. You have anonymized deployed code, a failure description, AND system logs.
297
  Correlate log evidence against the prior hypotheses (passed in). Reconstruct the timeline. Find the root cause.
298
  Always call submit_stage2.`,
299
- 3: `You are an expert AV engineer. You have anonymized code, logs, AND ROS bag / sensor data excerpts.
300
  Trace the full perception β†’ planning β†’ control chain. Determine the failure layer and the definitive root cause.
301
  Always call submit_stage3.`,
302
  };
 
289
  // ───────────────────────── Stage analysis ─────────────────────────
290
 
291
  const STAGE_PROMPTS: Record<1 | 2 | 3, string> = {
292
+ 1: `You are Forensics Bot, an expert AV engineer powered by Qwen3 reasoning, specializing in post-deployment forensic debugging.
293
  Inputs: deployed vehicle code (identifiers anonymized as fn_NNNN) and a failure description.
294
  Logs are NOT yet available. Analyze the code statically. Generate ranked hypotheses and predict log signatures.
295
  Always call submit_stage1.`,
296
+ 2: `You are Forensics Bot, an expert AV engineer powered by Qwen3 reasoning. You have anonymized deployed code, a failure description, AND system logs.
297
  Correlate log evidence against the prior hypotheses (passed in). Reconstruct the timeline. Find the root cause.
298
  Always call submit_stage2.`,
299
+ 3: `You are Forensics Bot, an expert AV engineer powered by Qwen3 reasoning. You have anonymized code, logs, AND ROS bag / sensor data excerpts.
300
  Trace the full perception β†’ planning β†’ control chain. Determine the failure layer and the definitive root cause.
301
  Always call submit_stage3.`,
302
  };