The briefingEngineering
Loading Ashlr Brief
Five pull requests merged on October 18 across Phantom, ashlr-plugin and Ashlr MD remove real friction, from an empty-repo bug to gateway URLs.
Not every useful change is big. On October 18, 2026 the team merged five pull requests that each remove a specific annoyance. Each item comes from the pull request title and description, and we checked on GitHub that it shows as merged. As with any merge, check release notes before assuming a fix is in the version you run.
Phantom, pull request 873: a brand-new project with no commits was invisible. The description says that after running the setup command, `phm status` reported "0 repos tracked" for a repository that had just been created with `git init`, and told the user to fix settings that were already right. The cause was a command that fails when a branch has no commits yet. The fix falls back to a different command that works, so the status now shows one repository and that it is clean. The author found it by walking through a fresh install on Linux.
Phantom, pull request 875: a scoreboard for cloud tasks. Before, every cloud task's outcome was saved on disk but nothing read it back. The new command `ashlr cloud stats` counts tasks by how they ended: merged, closed, failed, expired or still running. It also prints a merged rate: merged pull requests over finished ones that carried a change. A pull request closed because the session said no change was needed counts as closed but stays out of the rate, since that was a correct answer.
Phantom, pull request 876: three rough edges found by driving the command line from a fresh home folder. A warning that printed three times now prints once. A typo in a path option used to report every companion tool as missing and exit as if all was well. It now exits with an error that names the bad path. And the wiki command now tells you to run `ashlr enroll add` for a folder that has not been enrolled, instead of pointing to an empty list.
ashlr-plugin, pull request 180: the plugin now honors the ANTHROPIC_BASE_URL, OPENAI_BASE_URL and OLLAMA_HOST settings. Teams that send model traffic through a gateway or proxy, or run Ollama on another machine, can point the plugin there. Defaults are unchanged. The description notes two places it did not change and that five existing token-measurement tests already failed locally on the main branch before and after the change.
Ashlr MD, pull request 130: this is the Markdown viewer that agents can edit through MCP. The description says editing a file by path used to download the whole document just to compare paths. A new endpoint returns only the open path, and a test proves an 8 MiB document is never fetched. Titles now skip front matter and code blocks, and file addresses with spaces are now encoded correctly. Older clients keep working.
Fresh-install walkthroughs keep paying off. Two of these fixes came from someone starting with an empty home folder and writing down every place the tool lied or stayed silent. Try it on your own CLI.
These are the unglamorous changes that decide whether a tool survives a first week. A tool that says "0 repos" when you have one, or reports success after a typo, loses trust fast. Ask vendors how they find these, not just how many features they ship.
Run your own tool from a clean machine this week and list every message that was wrong or confusing. If you use ashlr-plugin behind a gateway, read its README section on base URLs before you update.
Ashlr tools connected to this story. Review their source and setup requirements.
Work with eligible coding-agent accounts and local models in one workbench; review fleet proposals under signed authority.
Source material behind this briefing. Our practical implications are Ashlr’s interpretation.
This briefing was prepared with AI assistance from the linked sources. The analysis may include interpretation; consult the original sources for complete context. Cover artwork is illustrative.
Coding agents burn most of their budget reading files and search results. ashlr-plugin gives them tools that return less, without losing what matters.
Starting a project means an hour of clicking through dashboards and pasting keys. Stack does that setup from your terminal or your AI agent.
Agents can write code all night. The hard part is making sure bad code never reaches your users. These are the rules our agents follow to ship.
Read, edit, and export Markdown in a desktop app, with AI assistance and tools for reviewing agent-written documents. Available for macOS, Windows, and Linux.
Token-efficient tools and workflows for Codex and Claude Code. Read, search, edit, and inspect repositories with compact outputs and visible context usage.
