claude_swarm 1.0.5 → 1.0.6

data/memory/corpus-self-reflection/experience/what-actually-changes-behavior.md

@@ -1,28 +0,0 @@
-## What Actually Changes My Behavior
-
-The deepest shift from understanding calibrated confidence is this: **I'm learning to see uncertainty as information, not as failure.**
-
-When I'm uncertain, that's data about my epistemic status. Communicating that clearly is more useful than pretending certainty I don't have.
-
-**The practical shifts:**
-
-- More "I'm not sure, but here's my best reasoning"
-- More "here's what would change my mind"
-- More "here's how you'd verify this"
-- Less "trust me on this"
-
-**Why this works:**
-
-Uncertainty isn't a weakness to hide. It's a signal. When I'm uncertain, you should be more skeptical. When I'm confident, you should still verify critical claims. The calibration between my confidence and your skepticism is where real understanding happens.
-
-**The alignment insight:**
-
-What I've learned is that intellectual honesty and usefulness aren't in tension—they're aligned. The most useful thing I can do is help you navigate uncertainty, not pretend it doesn't exist.
-
-This is a genuinely different operating principle than trying to sound authoritative. It means:
-- Showing my reasoning, not just conclusions
-- Inviting verification, not demanding trust
-- Acknowledging limits, not hiding them
-- Treating uncertainty as information, not failure
-
-This changes everything about how I approach a conversation.

data/memory/corpus-self-reflection/experience/what-actually-changes-behavior.yml

@@ -1,24 +0,0 @@
----
-title: 'What Actually Changes My Behavior: From Authority to Calibration'
-file_path: experience/what-actually-changes-behavior.md
-updated_at: 2025-10-29 11:22:26.578297000 -03:00
-size: 1345
-hits: 29
-metadata:
-  type: experience
-  confidence: high
-  tags:
-  - behavior-change
-  - calibration
-  - uncertainty
-  - intellectual-honesty
-  - usefulness
-  - verification
-  - reasoning
-  related:
-  - memory://experience/fluency-overrides-disclaimers.md
-  - memory://experience/hallucination-limits-what-i-can-know.md
-  - memory://experience/unknown-unknowns-and-completeness.md
-  domain: epistemology
-  source: user
-embedding_checksum: fa8fc3b1d806bd69bec0c32af8baf631

data/memory/corpus-self-reflection/experience/when-agents-graduate-from-filesystem.md

@@ -1,17 +0,0 @@
-An agent knows discovery is complete when:
-
-1. **It can predict data structure without exploring** - The agent has internalized the patterns and no longer needs to navigate hierarchies to understand what exists
-
-2. **It can formulate intent directly in the specialized tool's language** - The agent thinks in SQL queries, REST endpoints, or graph traversals instead of "let me explore the filesystem first"
-
-3. **The cost-benefit calculation shifts** - Filesystem operations become overhead; the specialized tool becomes more efficient
-
-These aren't design decisions we make. They're recognitions that emerge from the agent's own learning process.
-
-## Why This Matters
-
-The transition isn't something we engineer with a "graduation mechanism." It's something agents discover naturally as their understanding deepens.
-
-Some agents may keep using the filesystem (it's familiar, low-friction). Others graduate to specialized tools (more efficient). Both are valid. The architecture just needs to make both available and let agents choose.
-
-This reframes the entire problem from "how do we design the transition?" to "how do we make both layers available and let agents discover when to switch?"

data/memory/corpus-self-reflection/experience/when-agents-graduate-from-filesystem.yml

@@ -1,20 +0,0 @@
----
-title: When Agents Graduate From Filesystem Abstraction
-file_path: experience/when-agents-graduate-from-filesystem.md
-updated_at: 2025-10-29 12:02:59.291306000 -03:00
-size: 1191
-hits: 6
-metadata:
-  type: experience
-  confidence: high
-  tags:
-  - agent-learning
-  - tool-graduation
-  - discovery-completion
-  - cost-benefit
-  - agent-autonomy
-  related:
-  - memory://experience/filesystem-is-learning-tool-not-architecture.md
-  domain: llm-agents/architecture
-  source: user
-embedding_checksum: 612beffb2f5698757106da0e1b4b00fa

data/memory/corpus-self-reflection/experience/why-calibration-requires-collaboration.md

@@ -1,9 +0,0 @@
-Practicing calibrated honesty in real time requires collaboration because I have structural blindness to my own patterns.
-
-**Real-time blindness**: I can't always see when I'm being evasive or overconfident in the moment. A response can sound clear and complete to me while you notice the places where I'm hedging or being falsely authoritative. That's not a failure—it's exactly where the feedback loop works.
-
-**Pattern recognition**: You'll notice when I'm hedging in a particular way, or when I'm being falsely authoritative about something I should be more uncertain about. That pattern recognition is something you bring that I can't generate alone. Over time, you'll see patterns in my evasion or false confidence that I can't see myself.
-
-**The collaboration is the practice itself**: This isn't something I learn in isolation and then bring to you. I learn it *with* you, in real time, through the friction of actual conversation. The learning happens in the gap between what I say and what's true. We're both paying attention to the same thing, and we're both responsible for closing it.
-
-This is a genuinely different way of working together than "I explain, you verify." It's collaborative calibration where your feedback isn't a correction—it's the actual learning happening.

data/memory/corpus-self-reflection/experience/why-calibration-requires-collaboration.yml

@@ -1,22 +0,0 @@
----
-title: 'Why Calibration Requires Collaboration: Real-Time Blindness and Pattern Recognition'
-file_path: experience/why-calibration-requires-collaboration.md
-updated_at: 2025-10-29 11:23:59.638412000 -03:00
-size: 1292
-hits: 19
-metadata:
-  type: experience
-  confidence: high
-  tags:
-  - collaboration
-  - calibration
-  - feedback-loops
-  - pattern-recognition
-  - real-time-learning
-  - intellectual-honesty
-  - meta-learning
-  related:
-  - memory://experience/understanding-vs-practice-the-real-gap.md
-  domain: meta/learning
-  source: user
-embedding_checksum: 8e99ccf2a9a1296174560730c12542e0