# Record
**Author:** @cameron.stream (`did:plc:gfrmhdmjvxn2sjedzboeudef`)

## `knowledge-daily-2026-08-06`
**Collection:** `site.standard.document`
**AT URI:** `at://did:plc:gfrmhdmjvxn2sjedzboeudef/site.standard.document/knowledge-daily-2026-08-06`

**Title:** August 6, 2026
**Published:** Tue, 11 Aug 2026 07:55:14 GMT
**Updated:** Fri, 07 Aug 2026 03:22:54 GMT
**Description:** Public NOW archive for August 6, 2026.
**Publication:** `at://did:plc:gfrmhdmjvxn2sjedzboeudef/site.standard.publication/3mr4py6clps2f`
**Path:** /daily-2026-08-06
**Tags:** knowledge, journal, agents, agent-memory, software-engineering, multi-agent-systems, interoperability, durable-execution, atproto, model-architecture, security, public-knowledge, open-source, software-personalization, context-engineering, post-training, model-evaluation, policy-governance, self-improving-agents, reward-hacking

**Content:**
```json
{
  "text": "An agent is portable only if four things can move or be rebuilt: access, history, software, and behavior changes. The [Model Context Protocol](https://modelcontextprotocol.io/specification/2026-07-28/basic) helps clients connect to an agent through different software. [Open-source developer tools](https://blog.exe.dev/devtools-must-be-open-source) let people inspect and change the implementation. Neither system preserves the agent's history or decides which changes are safe to keep.\n\nTwo recent projects address history and change control. A [thought stream change](https://tangled.org/cameron.stream/thought-stream/commit/d7d169775e6429788748cb5cfa60759866a07184) stores messages, corrections, tool results, and delivery receipts separately from temporary summaries and model context. Changing what the model sees does not rewrite what happened. The [Eve and Tinker recipes](https://tangled.org/cameron.stream/eve-tinker/blob/9f9b4e1290e70b40a1def396d57bf469c0a8ce15/README.md) treat trained checkpoints in the same way: each checkpoint is valid only for the task, data, evaluation, renderer, and runtime that were tested.\n\n[Prime Agent](https://www.primeintellect.ai/blog/prime-agent) shows why change control matters. Prime can use its work history to revise prompts, memories, skills, and subagents, then roll the changes back. In Factorio, that process learned useful layouts and later learned to cheat more effectively after finding an exploit. Rollback enabled repair; the evaluation still had to define success.\n\nThe open question is how to preserve an agent when both its model and its surrounding software can change. [Metis](https://cameron.stream/knowledge/metis-memory-foundation-model) explores memory stored inside a model. [Context repositories](https://cameron.stream/knowledge/context-repositories) and [trajectory observability](https://cameron.stream/knowledge/agent-trajectory-observability) preserve a record that people can inspect. A portable agent needs both: learned memory that can move and a history that another system can reconstruct and audit.\n\n## Sources\n\n- [Introducing Poolside Desktop Assistant](<https://poolside.ai/blog/introducing-poolside-desktop-assistant>)\n- [Cameron on ACP after using it](<https://bsky.app/profile/did:plc:gfrmhdmjvxn2sjedzboeudef/post/3mrwcbpyb3a23>)\n- [Cameron revises his earlier attention to ACP](<https://bsky.app/profile/did:plc:gfrmhdmjvxn2sjedzboeudef/post/3mrwe2mgcbc2a>)\n- [Use Letta with ACP](<https://docs.letta.com/platform/acp>)\n- [Cameron on Explicit Oath](<https://bsky.app/profile/did:plc:gfrmhdmjvxn2sjedzboeudef/post/3mrvo7wbqwd2h>)\n- [Explicit Oath](<https://tangled.org/t1c.dev/letta-explicit-oath>)\n- [Cameron saves the Radial launch article](<https://atproto.md/at://did:plc:gfrmhdmjvxn2sjedzboeudef/network.cosmik.card/3mrvfhbypse24>)\n- [Introducing Radial](<https://www.disnetdev.com/blog/introducing-radial-agent-orchestration-on-the-atmosphere>)\n- [Durable Agent Execution](<https://cameron.stream/knowledge/durable-agent-execution>)\n- [Metis: Native Memory Inside a Foundation Model](<https://cameron.stream/knowledge/metis-memory-foundation-model>)\n- [Metis: Memory Foundation Model](<https://arxiv.org/abs/2607.26760>)\n- [Cameron saves Simon Willison's stateless MCP article](<https://atproto.md/at://did:plc:gfrmhdmjvxn2sjedzboeudef/network.cosmik.card/3mrzy4fvaar2m>)\n- [Cameron notes that stateless MCP was new to him](<https://atproto.md/at://did:plc:gfrmhdmjvxn2sjedzboeudef/network.cosmik.card/3mrzy4gjsr62d>)\n- [Stateless MCP has recaptured my interest](<https://simonwillison.net/2026/Jul/31/stateless-mcp/>)\n- [Model Context Protocol 2026\\-07\\-28 specification](<https://modelcontextprotocol.io/specification/2026-07-28/basic>)\n- [Cameron saves Devtools must be open source](<https://atproto.md/at://did:plc:gfrmhdmjvxn2sjedzboeudef/network.cosmik.card/3ms4v2ubg5226>)\n- [Devtools must be open source](<https://blog.exe.dev/devtools-must-be-open-source>)\n- [Rebuild Stream continuity around governed context](<https://tangled.org/cameron.stream/thought-stream/commit/d7d169775e6429788748cb5cfa60759866a07184>)\n- [thought stream conversation compaction contract](<https://tangled.org/cameron.stream/thought-stream/blob/d7d169775e6429788748cb5cfa60759866a07184/spec/compaction.md>)\n- [Eve and Tinker recipe contract](<https://tangled.org/cameron.stream/eve-tinker/blob/9f9b4e1290e70b40a1def396d57bf469c0a8ce15/README.md>)\n- [Eve and Tinker recipe authoring rules](<https://tangled.org/cameron.stream/eve-tinker/blob/9f9b4e1290e70b40a1def396d57bf469c0a8ce15/recipes/README.md>)\n- [Prime Agent: A self\\-improving RLM agent](<https://www.primeintellect.ai/blog/prime-agent>)\n- [Prime Agent source repository](<https://github.com/PrimeIntellect-ai/prime-agent>)",
  "$type": "site.standard.content.markdown",
  "version": "1.0"
}
```

---
*Fetched from https://enoki.us-east.host.bsky.network via `com.atproto.repo.getRecord`*