If you already use Vale

Vale is a command you run on files. It knows nothing about an agent, a turn, or a reply, and it forgets each run when it exits. just-say-so adds the parts that need a harness. Your Vale setup stays the source of truth.

What the plugin adds

It runs Vale at the moments the harness offers, and feeds the result back to the model. After a write, the agent gets the alerts for the lines it added. Before a gh or MCP call publishes text, the call is checked with the text in hand. At the end of a reply, the reply itself is linted. Vale on its own never reaches the model.

It enforces at the right moment for each target. A file is gated at the end of the turn, when the agent can still fix it. A publish call is denied before it runs, because nothing can un-publish. A reply is counted and reported on the next prompt.

It carries the rules a linter cannot express. “Lead with the answer” is an instruction the model can understand, but it’s not a lint-able rule that something like Vale can catch. The plugin keeps these judgment rules in the model’s context across session starts, compaction, and subagents. It also reads your Vale config before each injection: a bullet backed by Vale rules that you’ve turned off is left out, and error-level rules of yours that no bullet describes are named at the end.

It remembers what you rejected. Vale has no record of the reply you found bad. /just-say-so:restate keeps it, with your reason and tag. The summary counts, and the advice turns recurring tags into Vale tunes, judgment bullets, and package suggestions.

What stays yours

• The same .vale.ini drives your editor, your CI, the commit gate, and every check here. The plugin reads the nearest one above the file it checks.
• The JustSaySo rules are an ordinary Vale package. Add it to Packages, run vale sync, and your other styles sit beside it.
• Your vocabulary is Vale’s accept.txt. /just-say-so:allow writes to it; you can edit it by hand.
• Severity is the policy. An error can block; a warning or suggestion is advice. Set levels in .vale.ini as you always have.

Two things to know

Chat replies lint under the name reply.reply, and published text under names such as gh.pr.body.publish, so one .vale.ini narrows rules per target with sections like [*.reply]. The Vale engine has the lint names and the resolution order.

The reminder the model gets is derived from your config, not fixed. /just-say-so:doctor prints the plugin’s reading of your .vale.ini next to the raw sections, so you can check the two agree.