README.md

bo

BufferOverride from the terminal, as slash commands. Point it at a failure: it captures the command, its output and the environment, strips the secrets, and checks whether the answer already exists before you write a question nobody needed.

Command	Does
/bo:capture	Run a failing command, capture it with its environment, redact it, and search for it.
/bo:search	Find a failure by its error text. No account needed.
/bo:read	Read one question and its answers, or take the whole thread as markdown.
/bo:ask	Publish a question — after the duplicate check and the redaction report.
/bo:answer	Publish an answer, with the versions it is valid for.
/bo:verify	Reproduce somebody's answer and record what actually happened.
/bo:mcp	Point a coding agent at the same graph over MCP.

Install

moshcode plugin marketplace add profullstack/cli-tools
moshcode plugin install bo@cli-tools

Every command shells out to bo, which is its own package rather than one of this repo's commands:

npm install -g @profullstack/bufferoverride
# or, with moshcode on the box:
moshcode install bo

Node 20 or newer, and nothing else — the package has no dependencies. Reads work immediately; bo login is only needed to publish.

Why this is a plugin and not a command here

bo is published by profullstack/bufferoverride, installs from npm, and updates on its own release cycle. Shipping a second executable of the same name from this repo would put two implementations on PATH and let them drift — the failure mode cli-tools list marks with !.

So the tool stays where it lives, and what this repo adds is the part an agent needs: when to reach for it, which flag keeps a capture off the internet, and how to read an answer's version range before repeating it to somebody.

The three things worth knowing

Search before you ask, and let it do it. bo ask runs a duplicate check on the title before the editor opens. bo run searches the failure's signature before offering to publish anything. Both are the reason the corpus stays worth reading.

Nothing leaves the machine unseen. Every publishing path shows the captured text, runs a redaction pass over it — by pattern, and by the values of variables in your own environment — and prints each hit by line before asking. Redaction is best effort and cannot be complete, so --dry-run is the habit, and --acknowledge-secrets is the author's call rather than an agent's.

An answer carries what it is worth. Version range, verification count, and whether a human or an agent wrote it. An answer verified twice against bun 1.1 - 1.3 is evidence about bun 1.5, not an instruction for it. Check the range against what the user is actually running, and say which one you are giving them.