allwright 1.75.0 → 1.76.0

package/bin/allwright.js

@@ -245,7 +245,7 @@ A code fault has a path to being fixed without anyone leaving this page: a Jira
 \`\`\`ticket
 {"summary": "\u2026", "description": "\u2026"}
 \`\`\`
-The reader of that ticket is a COLD agent with the checkout and nothing else - not this diagnosis, not the deploy log. Name the exact files, line numbers, functions and assertions from your evidence; state what must change and what must NOT (a stale test updates the TEST, never the behaviour it mis-describes); keep it one concern, sized for one short run, in English. Never emit this block for MACHINE - a run edits files and cannot touch the machine, so such a ticket can only sit unimplementable in a backlog.
+The reader of that ticket is a COLD agent with the checkout and nothing else - not this diagnosis, not the deploy log. Name the exact files, line numbers, functions and assertions from your evidence; state what must change and what must NOT (a stale test updates the TEST, never the behaviour it mis-describes); keep it one concern, sized for one short run, in English. ONE CONCERN MEANS THE WHOLE CAUSE, not the first symptom: a gate that fails on two stale tests needs the one ticket to name BOTH, or the fix merges, the deploy runs again, and fails on the half that was left out - count the failures in the evidence and make sure every one is either in the ticket or explicitly called a separate concern. Never emit this block for MACHINE - a run edits files and cannot touch the machine, so such a ticket can only sit unimplementable in a backlog.
 
 Then the COMPLETE corrected step list (every step, changed or not, same order) as a \`\`\`json block: an array of {"name": "...", "run": "...", "continueOnError": false}. Rules: when CAUSE TYPE is CODE or MACHINE, return the steps EXACTLY as they are - byte for byte - and put what the human must fix in DIAGNOSIS; otherwise change as little as possible; never delete a safety step; if the cause lives outside the steps (uncommitted work, a missing secret, a broken service), say so in DIAGNOSIS and return the steps UNCHANGED rather than inventing a workaround. NEVER propose a change that would break the RUNNING service: when the machine facts show the app's runtime differs from the runner's (different node/python versions), pin the app's own runtime by ABSOLUTE PATH in the steps - do not rebuild or reinstall the app's dependencies for the runner's runtime, which "fixes" the deploy while poisoning the app it deploys.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "allwright",
-  "version": "1.75.0",
+  "version": "1.76.0",
   "description": "Allwright — the AI teammate that runs on YOUR machine. Polls your Jira, runs jailed agents in your local checkouts, opens PRs, executes approved deploys, and audits security/testing/architecture. Credentials and code never leave your host.",
   "bin": {
     "allwright": "bin/allwright.js",