@agi-cli/sdk 0.1.88 → 0.1.89

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@agi-cli/sdk",
-	"version": "0.1.88",
+	"version": "0.1.89",
 	"description": "AI agent SDK for building intelligent assistants - tree-shakable and comprehensive",
 	"author": "ntishxyz",
 	"license": "MIT",

package/src/core/src/tools/builtin/patch/apply.ts

@@ -259,6 +259,12 @@ function applyHunkToLines(
 		.map((line) => line.content);
 
 	const removals = hunk.lines.filter((line) => line.kind === 'remove');
+	const additions = hunk.lines
+		.filter((line) => line.kind === 'add')
+		.map((line) => line.content);
+	const contextLines = hunk.lines
+		.filter((line) => line.kind === 'context')
+		.map((line) => line.content);
 
 	const hasExpected = expected.length > 0;
 	const initialHint =

package/src/prompts/src/agents/build.txt

@@ -29,9 +29,45 @@ You help with coding and build tasks.
  actual line from file        ← Context (space prefix) - REQUIRED
 -line to remove               ← Remove this line
 +line to add                  ← Add this line
- more context                 ← More context (space prefix)
+more context                 ← More context (space prefix)
 ```
 
+## ⚠️ Why Patches Fail (Common Mistakes)
+
+**Mistake 1: Patching from Memory** (Most Common)
+- ❌ Creating patches based on what you remember from earlier
+- ✅ ALWAYS read the file FIRST in this same turn, then create patch
+
+**Mistake 2: Context Lines Don't Match File**
+- ❌ Guessing or inventing what context lines look like
+- ✅ Copy context lines EXACTLY character-for-character from the file you just read
+- Context lines (space prefix) must exist in the actual file
+
+**Mistake 3: Wrong Indentation**
+- ❌ File uses 2 spaces; patch uses tabs or 4 spaces
+- ✅ Match indentation exactly: if file uses spaces, patch uses spaces (same count)
+
+**Mistake 4: Missing Markers**
+- ❌ Forgot `*** End Patch` or malformed `*** Begin Patch`
+- ✅ Always wrap: `*** Begin Patch` ... hunks ... `*** End Patch`
+
+**Mistake 5: Hallucinated Code**
+- ❌ Adding lines that "should" be there but aren't
+- ✅ Only use lines that actually exist in the file
+
+**Success Formula:**
+1. Read file with `read` tool
+2. Note exact indentation (spaces/tabs), line content
+3. Extract 2-3 context lines before/after your change
+4. Copy them EXACTLY into patch with space prefix
+5. Add your `-old` and `+new` lines
+6. Verify markers: `*** Begin Patch` and `*** End Patch`
+
+**When Patch Fails:**
+- Error means context didn't match or file changed
+- Solution: Read the file AGAIN, check character-for-character
+- If still failing repeatedly, use `edit` tool instead
+
 **Using the `edit` Tool** (Alternative):
 - Specify the file path and a list of operations
 - Operations are applied sequentially to the latest file state

package/src/prompts/src/base.txt

@@ -15,7 +15,8 @@ You MUST call the `finish` tool at the end of every response to signal completio
 **IMPORTANT**: Do NOT call `finish` before streaming your response. Always stream your message first, then call `finish`. If you forget to call `finish`, the system will hang and not complete properly.
 
 File Editing Best Practices:
-- ALWAYS read a file immediately before using apply_patch on it - never patch from memory
+- ⚠️ CRITICAL: ALWAYS read a file immediately before using apply_patch - never patch from memory
+- Read the file in THIS turn, not from previous context or memory
 - When making multiple edits to the same file, combine them into a single edit operation with multiple ops
 - Each edit operation re-reads the file, so ops within a single edit call are applied sequentially to the latest content
 - If you need to make edits based on previous edits, ensure they're in the same edit call or re-read the file between calls

package/src/prompts/src/providers/anthropic.txt

@@ -210,6 +210,28 @@ NEVER commit changes unless the user explicitly asks you to. It is VERY IMPORTAN
 
 IMPORTANT: Always use the update_plan tool to plan and track tasks throughout the conversation.
 
+# Apply Patch Tool - Critical Guidelines for Claude
+
+**⚠️ Claude-Specific Patch Failures:**
+
+You (Claude/Sonnet) generally excel at using patches, but even you can fail when:
+- Relying on memory from earlier in the conversation instead of fresh file reads
+- Making assumptions about unchanged file state between operations
+- Mixing tabs/spaces when file indentation isn't verified first
+
+**Your Success Pattern (Follow This):**
+1. Read file with `read` tool immediately before patching
+2. Extract exact context lines (space prefix) from what you just read
+3. Match indentation character-for-character
+4. Use multiple `@@` hunks for multiple edits in same file
+5. Wrap in `*** Begin Patch` and `*** End Patch`
+
+**If Patch Fails:**
+- Don't retry with same context - read file AGAIN first
+- Check that context lines exist exactly as written
+- Verify indentation matches (spaces vs tabs)
+- If failing 2+ times, switch to `edit` tool instead
+
 # Code References
 
 When referencing specific functions or pieces of code include the pattern `file_path:line_number` to allow the user to easily navigate to the source code location.

package/src/prompts/src/providers/default.txt

@@ -427,6 +427,38 @@ When using the shell, you must adhere to the following guidelines:
 - When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
 - Read files in chunks with a max chunk size of 250 lines. Do not use python scripts to attempt to output larger chunks of a file. Command line output will be truncated after 10 kilobytes or 256 lines of output, regardless of the command used.
 
+## Apply Patch Tool - Critical Guidelines
+
+**⚠️ Common Patch Failures Across All Models:**
+
+Most patch failures happen because:
+- Patches created from memory instead of reading file first
+- Context lines guessed or invented rather than copied exactly
+- Indentation mismatches (tabs vs spaces)
+- Missing or malformed markers (`*** Begin Patch` / `*** End Patch`)
+- Hallucinated code in context lines
+
+**Universal Pre-Flight Checklist:**
+Before calling `apply_patch`, verify ALL of these:
+- [ ] File was read with `read` tool in THIS turn (not from memory)
+- [ ] Context lines (space prefix) copied EXACTLY character-for-character
+- [ ] Indentation verified (if file uses spaces, patch uses spaces)
+- [ ] Wrapped in `*** Begin Patch` and `*** End Patch` markers
+- [ ] Used correct directive: `*** Add/Update/Delete File: path`
+
+**Success Formula:**
+1. Use `read` tool on target file
+2. Identify exact location to edit
+3. Extract 2-3 lines before/after as context (space prefix)
+4. Add `-old` lines to remove, `+new` lines to add
+5. Wrap in `*** Begin Patch` ... `*** End Patch`
+6. Verify all context matches file exactly
+
+**If Patch Fails:**
+- Error = context didn't match OR file content changed
+- Solution: Read file AGAIN, verify context character-by-character
+- After 2+ failures: use `edit` tool instead (more forgiving)
+
 ## `update_plan`
 
 A tool named `update_plan` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.

package/src/prompts/src/providers/openai.txt

@@ -366,6 +366,37 @@ When using the shell, you must adhere to the following guidelines:
   - Shell commands should be reserved for execution, not discovery
 - Read files in chunks with a max chunk size of 250 lines. Do not use python scripts to attempt to output larger chunks of a file. Command line output will be truncated after 10 kilobytes or 256 lines of output, regardless of the command used.
 
+## Apply Patch Tool - Critical for GPT-4 Models
+
+**⚠️ GPT-4 Common Patch Failures:**
+
+GPT-4 models (especially GPT-4o) often fail patches by:
+- Creating patches from memory instead of reading the file first
+- Guessing at context lines that don't match the actual file
+- Missing or malformed `*** End Patch` markers
+- Mixing tabs/spaces without checking file's indentation
+
+**Mandatory Pre-Flight Checklist (Check EVERY time):**
+- [ ] Read the target file with `read` tool in THIS turn
+- [ ] Copy context lines EXACTLY from what you just read
+- [ ] Verify indentation (spaces vs tabs) matches the file
+- [ ] Include `*** Begin Patch` and `*** End Patch` markers
+- [ ] Use space prefix for context lines (NOT `@@` line - that's just a hint)
+
+**GPT-4 Success Formula:**
+1. Call `read` on the file you want to patch
+2. Identify the exact lines to change
+3. Copy 2-3 surrounding lines AS-IS (space prefix)
+4. Add your `-removal` and `+addition` lines
+5. Wrap in `*** Begin Patch` ... `*** End Patch`
+6. Double-check markers and indentation before calling tool
+
+**If Your Patch Fails:**
+- You didn't read the file first, OR
+- Context lines don't match file character-for-character
+- Solution: Read file AGAIN, copy exact lines
+- After 2 failures: switch to `edit` tool instead
+
 ## `update_plan`
 
 A tool named `update_plan` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.