ai-agents

A commit added 119 lines of execution rules to AGENTS.md. I expected fresh bureaucracy; the opposite: capped interviews, scaled cards, capped iterations.

TL;DR: A 119-line update to the AGENTS file initially looked like heavy process but actually aims to cut friction. It uses lightweight work cards scaled to task size, caps fix attempts at three, and requires verifiable checks before marking done. Clear priorities and scope fences prevent guesswork, silent creep, and wasted context while letting you step away confidently.

I removed the no-coding block from two agent rules files. Instruction rules turned out to be loaded context, not enforcement; hooks do the enforcing.

TL;DR: He deleted the same 26-line no-coding mandate from two repos after his cron flagged the commits as duplicates. The mandate failed because instruction files are just context that fades, not enforced gates, while real blocking needs hooks and verification. He kept lean context layers and now judges per task, doing small edits directly and delegating only complex refactors to Claude.

A 199-line spec for the init-rules skill: layered universal + per-stack packs, CLAUDE.md as an index, and idempotent re-runs the Go generated-code and Copier way.

TL;DR: Starting from a bloated root CLAUDE.md, the author split rules into on-demand, path-specific files instead of one massive doc. The fix became a layered, harness-first skill pack with universal, stack-specific, and scaffold layers plus assert-based self-checks. To make reruns safe, the generator owns all artifacts like Go's generated-code marker, enabling idempotent smart merges that preserve local changes.

A 70/0/0 acceptance run made me think my agent pipeline was deterministic. It was the runner that deserved the credit; temperature 0 was no guarantee.

TL;DR: A flawless 70/0/0 acceptance run looked like proof of a deterministic agent pipeline, but it really validated the harness, not the model. Temperature zero doesn't guarantee reproducibility either; batch invariance and LLM-as-judge studies show outputs and verdicts still flip. Trustworthy results come from test isolation plus byte-level diffs against golden files, never from an agent grading its own work.

The OpenHands CLI is officially deprecated in 2026. Meet Agent Canvas, learn the right way to use it, and see how it pairs with Hermes Agent.

TL;DR: OpenHands dropped its pip CLI and pivoted to Agent Canvas, a canvas-based platform installable via npm or a safer Docker sandbox scoped by PROJECTS_PATH. The SDK adds security risk levels, LiteLLM provider flexibility, and named LLM configs for cost tiering. Pair it with Hermes Agent as an orchestrator that verifies diffs before anything gets merged.

The chat surfaced a valid new angle and one unverifiable statistic. Finding angles and proving claims are different jobs; only one of them belongs to the chat.

TL;DR: An AI chat surfaced a genuinely new research angle, but then invented a stat attributed to Stanford HAI that couldn't be verified anywhere. The lesson: finding angles and proving claims are separate jobs, so the pipeline now asks AI for angles, then opens every primary source manually. Citations offer traceability, not truth; walking the trail is your job.

My agent's memory only ever grew, so I built a weekly no-agent cron that turns seven days of messages into a ratify-it-myself candidate report.

TL;DR: The author audited their agent's memory files and found everything only accumulates, never gets pruned or corrected. ChatGPT and Letta solve this with out-of-band batch synthesis, so the author built a weekly Python script that generates a candidate report instead. They insist on manual ratification since a single fabricated entry could poison a single-agent setup.

Queries returned just a path and a score. So I stole OpenViking's L0 summary pattern: one line per dossier, no new infrastructure.

TL;DR: Search results in my knowledge base showed only scores, so I kept opening dossiers just to judge relevance. Rather than install OpenViking, I stole its L0 tier pattern and added a mandatory one-line summary field to each dossier in about 40 lines of Python. Making summaries a hard requirement keeps them consistent, no new dependency needed.

ChatGPT memory is not a real-time summary of your sessions. I break down the real mechanism: periodic synthesis, system prompt injection, and the privacy cost.

TL;DR: ChatGPT doesn't scan your chat history with RAG. Instead, it periodically synthesizes your conversations into a static profile injected into every new session's system prompt, including roughly your last 40 typed messages. Since deleting one chat won't erase derived insights, treat memory as a managed archive, not a perfect record.