DEPRECATED — redirects to pair-build. Use /sdlc:pair-build directly.
This skill redirects to /sdlc:pair-build. Invoke it now:
Skill("sdlc:pair-build", args="$ARGUMENTS")
Do NOT continue with the old build workflow below. Invoke pair-build and follow its instructions instead.
Log skill invocation:
Use $PLUGIN_DIR (already detected in Step 1 via find . -name "run-gates.sh"):
bash "$PLUGIN_DIR/../scripts/audit-trail.sh" log build sdlc:build started --context "$ARGUMENTS"bash "$PLUGIN_DIR/../scripts/audit-trail.sh" log build sdlc:build completed --context="<summary>"Correctness is the only measure that matters. Speed is not a factor. Not a preference. Not a tiebreaker. If a gate takes 10 minutes to run, you run it. If a checkpoint reveals drift, you re-run everything from scratch. If coverage is 94.9%, you write more tests. There are no "close enough" results.
Every quality claim in this skill is proven by an executable script that writes machine-readable proof to .quality/proof/. Claude's context window is ephemeral — proof files are not. If you did not run the script, you did not prove anything.
Never leave a dirty tree. Every phase transition, every plan item completion, every fix — commit immediately with conventional commit format. Do not accumulate uncommitted changes. Do not ask the user what to do with uncommitted code. Do not wait for a "good stopping point." The stopping point is NOW.