fix(release): v2.9.3 - foreground save, fix hallucinated user messages
CRITICAL: run_in_background callbacks caused model to re-engage after save, hallucinate fake "Human:" messages, and generate unsolicited multi-paragraph responses. Switch to foreground cat > heredoc which executes sub-second with no callback. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
@@ -30,7 +30,7 @@ metadata:
|
||||
- prompts
|
||||
---
|
||||
|
||||
# last30days v2.9.2: Research Any Topic from the Last 30 Days
|
||||
# last30days v2.9.3: Research Any Topic from the Last 30 Days
|
||||
|
||||
Research ANY topic across Reddit, X, YouTube, TikTok, Hacker News, Polymarket, and the web. Surface what people are actually discussing, recommending, betting on, and debating right now.
|
||||
|
||||
@@ -513,7 +513,7 @@ After displaying the invitation, save the complete research briefing to `~/Docum
|
||||
📎 ~/Documents/Last30Days/{slug}.md
|
||||
```
|
||||
|
||||
**Then immediately save using a background Bash command** (`run_in_background: true`):
|
||||
**Then save using a foreground Bash command** (cat is instant - no delay):
|
||||
|
||||
```bash
|
||||
mkdir -p ~/Documents/Last30Days && cat > ~/Documents/Last30Days/{slug}.md << 'RESEARCH_EOF'
|
||||
@@ -534,22 +534,22 @@ mkdir -p ~/Documents/Last30Days && cat > ~/Documents/Last30Days/{slug}.md << 'RE
|
||||
{The 2-3 specific suggestions from the invitation block}
|
||||
|
||||
---
|
||||
*Generated by [last30days](https://github.com/mvanhorn/last30days-skill) v2.9.2*
|
||||
*Generated by [last30days](https://github.com/mvanhorn/last30days-skill) v2.9.3*
|
||||
RESEARCH_EOF
|
||||
```
|
||||
|
||||
**CRITICAL RULES:**
|
||||
1. NEVER use the `Write` tool — it displays "Wrote N lines..." which ruins the experience
|
||||
2. ALWAYS use `run_in_background: true` so the Bash call is nearly invisible
|
||||
3. The `📎` line is part of your text message, not a separate tool call
|
||||
4. The invitation + `📎` line must be the LAST visible thing on screen
|
||||
5. When the background save completes, do NOT output any text. No "saved", no "ready for questions", nothing. Silently ignore the completion notification.
|
||||
1. NEVER use the `Write` tool - it displays "Wrote N lines..." which clutters the output
|
||||
2. NEVER use `run_in_background: true` - background callbacks cause the model to re-engage, hallucinate fake user messages, and generate unsolicited responses
|
||||
3. Use FOREGROUND Bash with `cat >` - it executes instantly (sub-second) and produces no output
|
||||
4. The `📎` line is part of your text message, not a separate tool call
|
||||
5. Do NOT output any text after the save Bash call. No "saved", no "ready for questions", nothing.
|
||||
|
||||
---
|
||||
|
||||
## WAIT FOR USER'S RESPONSE
|
||||
|
||||
**STOP and wait** for the user to respond. When the background save command completes, do NOT acknowledge it - just continue waiting silently.
|
||||
**STOP and wait** for the user to respond. Do NOT generate any text after the save.
|
||||
|
||||
---
|
||||
|
||||
|
||||
Reference in New Issue
Block a user