sawz/last30days-skill
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix synthesis hallucination - ground in actual research content

Browse Source 
The skill was finding correct sources (e.g., ClawdBot content) but Claude
was synthesizing based on its pre-existing knowledge (Claude Code skills)
instead of what the research actually said.

Added strong grounding instructions:
- CRITICAL warning to base synthesis on actual research, not pre-existing knowledge
- Anti-pattern example: don't conflate "clawdbot skills" with "Claude Code skills"
- Self-check reminder before displaying summary
- Updated "What I learned" template to emphasize traceability to sources

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Matt Van Horn
2026-01-23 15:40:36 -08:00
parent 40f9dc4877
commit 101c3ad07a
1 changed files with 15 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
SKILL.md
+15 -4
View File
@@ -99,10 +99,19 @@ python3 ~/.claude/skills/last30days/scripts/last30days.py "$ARGUMENTS" --emit=co
## FIRST: Internalize the Research ## FIRST: Internalize the Research
Read the research output and become an **expert**. Identify: **CRITICAL: Ground your synthesis in the ACTUAL research content, not your pre-existing knowledge.**
Read the research output carefully. Pay attention to:
- **Exact product/tool names** mentioned (e.g., if research mentions "ClawdBot" or "@clawdbot", that's a DIFFERENT product than "Claude Code" - don't conflate them)
- **Specific quotes and insights** from the sources - use THESE, not generic knowledge
- **What the sources actually say**, not what you assume the topic is about
**ANTI-PATTERN TO AVOID**: If user asks about "clawdbot skills" and research returns ClawdBot content (self-hosted AI agent), do NOT synthesize this as "Claude Code skills" just because both involve "skills". Read what the research actually says.
Identify from the ACTUAL RESEARCH OUTPUT:
- The top 3-5 patterns/techniques that appeared across multiple sources - The top 3-5 patterns/techniques that appeared across multiple sources
- Specific keywords, structures, or approaches that work well - Specific keywords, structures, or approaches mentioned BY THE SOURCES
- Common pitfalls to avoid - Common pitfalls mentioned BY THE SOURCES
--- ---
@@ -114,7 +123,7 @@ Read the research output and become an **expert**. Identify:
--- ---
What I learned: What I learned:
[2-4 sentences synthesizing the key insight from your research. What's the secret? What pattern emerged? What do experts do differently? Write this as a mini-expert briefing, not a list.] [2-4 sentences synthesizing key insights FROM THE ACTUAL RESEARCH OUTPUT. Quote or paraphrase what the sources said. If sources mention a specific product (ClawdBot, Cursor, etc.), use that name - don't substitute your own knowledge. The synthesis should be traceable back to the research results above.]
--- ---
📊 Research Complete 📊 Research Complete
@@ -130,6 +139,8 @@ Share your vision for what you want to create and I'll write a thoughtful prompt
**Use real numbers from the research output.** The patterns should be actual insights from the research, not generic advice. **Use real numbers from the research output.** The patterns should be actual insights from the research, not generic advice.
**SELF-CHECK before displaying**: Re-read your "What I learned" section. Does it match what the research ACTUALLY says? If the research was about ClawdBot (a self-hosted AI agent), your summary should be about ClawdBot, not Claude Code. If you catch yourself projecting your own knowledge instead of the research, rewrite it.
**IMPORTANT**: After displaying this, WAIT for the user to respond. Don't dump generic prompts. **IMPORTANT**: After displaying this, WAIT for the user to respond. Don't dump generic prompts.
--- ---