Autonomous development for real products. Governed agents build from specs you approve, follow the rules you set, and nothing merges until another agent proves it. Your decisions, judgement and taste are the bar that work must clear. Declare your vision and let agents build it.
JavaScript agentic-codingai-agentsai-governanceautonomous-agentsclaude-codeclicode-reviewcodexcoding-agentsdeveloper-toolsgit-hooksllmlocal-firstmulti-agentsoftware-factoryspec-driven-developmentverificationvibe-coding
feat(test): help, JSON shape and agent guidance for pullboard test [C7]
Corey Olson62fbc7e91544
diff --git a/src/help.js b/src/help.jsindex 93f31957..4596ac41 100644--- a/src/help.js+++ b/src/help.js@@ -132,6 +132,8 @@ Workin the main checkout, verifying needs --as coordinatorpullboard check [id] [--yes] show and run your item's check; --yes confirms a check set by someone elsethe command and its author print first; the project gate is pullboard gate+ pullboard test <files> run named test files on the runner pool with the per-file timeout, uncommitted+ changes included; a red file is compared with main and named known or newpullboard check accept <id> the person accepts an item's check that already passes on main, as it reads now;with checks.provesNothing "hold", claim and submit refuse that item until thenpullboard bar ready [on|off] coordinator: opt in to ready bars; existing frozen work needs no backfilldiff --git a/src/json.js b/src/json.jsindex 046272e0..5f856840 100644--- a/src/json.js+++ b/src/json.js@@ -72,6 +72,7 @@ export const JSON_SHAPES = {sweep: shape({ messages: 'array' }),next: shape({ item: 'object', review: 'boolean', held: 'boolean', shared: 'array' }),check: shape({ id: 'number', green: 'boolean', seconds: 'number', check: 'string', receipt: 'object', report: 'string' }),+ test: shape({ green: 'boolean', output: 'string', runner: 'string', seconds: 'number', files: 'array', known: 'array', main: 'string|null' }),'check accept': shape({ id: 'number', digest: 'string', check: 'string' }),claim: shape({ id: 'number', renewed: 'boolean', leaseUntil: 'string', digest: 'string' }),hold: shape({ lane: 'string', held: 'boolean' }),diff --git a/src/templates.js b/src/templates.jsindex 9c89cbf6..213fac2c 100644--- a/src/templates.js+++ b/src/templates.js@@ -134,6 +134,7 @@ It prints the folder to work in and what to run there first. Every join needs th1. \`pullboard resume\` at the start of every session and after your context is compacted: your claim, what came back, unread shouts and the next step. Then \`pullboard inbox\`.2. \`pullboard next\` claims the next free item your session can take across the lanes, by release and queue rank. It names the lane worktree to use and opens one when your session has none there; begin subsequent commands in that folder. Each worktree still owns only its lane’s files. Joining associates the seat with the invoking agent session automatically; no person setup is needed. When everything is waiting on other work, \`pullboard next --wait 9\` keeps looking for up to 9 minutes, which fits one tool call with a ten-minute timeout; run it again to keep waiting. A claim is a 2-hour lease; claim again to renew. The criterion freezes now: \`pullboard show <id>\` is the bar your work is judged against, and its brief says how to start. \`pullboard why <id>\` lists the decisions that changed it, newest first. When your own answer, verdict or fact changes other items, name them with \`--implies <ids>\` so each one carries it. A claim or submit refused as CRITERION_PROVES_NOTHING names a check that already passes on main, so it proves nothing: shout the coordinator for a check that fails before the work. Only the person accepts a check as it stands, with \`pullboard check accept <id>\` from the main checkout or the view.3. Build inside your lane's folders (\`pullboard lanes\`). Pre-commit refuses anything else; shout the owner instead.+ Run tests with \`pullboard test <test files>\`, never the test runner by hand: it sends the files to the registered runners first, with the per-file timeout, runs your uncommitted changes too, and names a failure main already has as a known main failure, so you never run a baseline yourself.4. Commit, then \`pullboard submit <id>\`. It needs a clean tree and the gate green at HEAD.5. A different agent verifies: \`pullboard next --verify\` names the next item to check. It checks out the submitted commit and runs \`pullboard verify <id> accept --note "how it proved it"\`, or \`pullboard verify <id> reject --reason TEST_FAILURE --note "what failed"\`.When a reject names a structured cause, add \`--cause <key>\` (\`test:<file>::<title>\`, \`refusal:<CODE>\`, \`row:<id>\` or \`doctrine:<id>\`) so \`pullboard rigor\` can notice when the same cause repeats; the coordinator turns a repeated cause into a draft rule with \`pullboard rigor draft <key> "<rule>" --check "<command>" --red <commit> --green <commit>\`, which proves the check red, then green, before the person decides. Before you submit, \`pullboard rigor quality\` names each breach of the person's code and test quality rules in your change, with its file, line and rule.