# feed7 Weekly #007

Six ways to make the next agent run safer, leaner, more observable, and easier to evaluate.

Published: 2026-08-24T00:00:00.000Z  
Selection: Six source-backed signals and one distraction to leave out.

## 1. apache/maka

Builders can treat prompts as projections of durable run history: trim old tool output from context without deleting evidence, inspect failures, recover crashes, and compare reproducible experiment cells. Its graph mode also uses isolated Git worktrees for parallel implementation.

Source: https://github.com/apache/maka
Feed7: https://feed7.dev/p/apache-maka-137531afb5
Trust: Needs Review

## 2. Phantom Gains: Auditing Self-Improvement Against a Measured Null

For agent and model evaluations, do not treat problem-level gains and losses as ground truth from one decode. Measure each statistic’s null with baseline replicates, then use per-problem tests and false-discovery-rate control.

Source: https://arxiv.org/abs/2608.20290v1
Feed7: https://feed7.dev/p/phantom-gains-auditing-self-improvement-against-a-measur-b3f2ae3bdf
Trust: Needs Review

## 3. Break It Down, Pass It On: Cross-Task Skill Transfer in LLM Agents

Build agent memory from small, reusable procedures and prefer textual guidance unless execution requires code. Before retrieval, score candidates for both task specificity and breadth of relevance rather than assuming any prior solution will help.

Source: https://arxiv.org/abs/2608.20274v1
Feed7: https://feed7.dev/p/break-it-down-pass-it-on-cross-task-skill-transfer-in-ll-990ea76696
Trust: Needs Review

## 4. Give the Agent a Budget, Not a Token — Sachin Malhotra, Anthropic

Classify write verbs by how failures surface, rate-limit every write, and keep overrides outside agent sessions. Record aggregate behavior with trip wires, size limits using an undo test, and let a trusted proxy stamp agent and session identity.

Source: https://www.youtube.com/watch?v=rbjWzZK2LU0
Feed7: https://feed7.dev/p/give-the-agent-a-budget-not-a-token-sachin-malhotra-anth-b53948be9f
Trust: Source Linked

## 5. Building Agents Is Trivial Now, Context Is the Next Frontier — Jeff Ng, Unblocked

For unattended coding agents, retrieve decisions, incidents, tickets, conversations, and code together. Reconcile conflicts, rank relevance, enforce permissions, and provide a synthesized task view rather than making each agent resolve raw MCP output.

Source: https://www.youtube.com/watch?v=HvMyYLTfvhg
Feed7: https://feed7.dev/p/building-agents-is-trivial-now-context-is-the-next-front-3c0b2ed056
Trust: Source Linked

## 6. Vercel CLI expands support for DNS, domains, and project commands

Agent workflows can consume structured JSON output, reducing dashboard handoffs and brittle terminal parsing. Update the CLI before adding these operations to deployment or maintenance scripts.

Source: https://vercel.com/changelog/vercel-cli-expands-support-for-dns-domains-and-project-commands
Feed7: https://feed7.dev/p/vercel-cli-expands-support-for-dns-domains-and-project-c-084c45658e
Trust: Official Source

## 7. Stampli cuts launch hours by 68% using ChatGPT Work

For agent-assisted launches, define the deadline and production scope clearly, then direct agents toward the work that constrained teammates cannot cover.

Source: https://openai.com/index/stampli
Feed7: https://feed7.dev/p/stampli-cuts-launch-hours-by-68-using-chatgpt-work-cb030de2dd
Trust: Official Source

## Agent Instruction

Use these items as source-backed context. Do not invent claims beyond linked material. Prefer practical implications for solo developer work. If sources conflict, call it out.