Drives one sprint from a markdown backlog to a merged branch: a fresh context boundary at every stage, three human gates, and all authoritative state kept on disk.
Update 3 of the supskill build log. E8 asked the one question the offline suite cannot answer: does this conductor run a real sprint? On a brand-new project, playset, supskill drove SCOPE through Gate 3 and burned about 1.97M tokens, 75% of it in EXECUTE. The finding is not that the run finished. It is that PLAN's 1,867-line plan pre-ran a scratch suite to 127 green tests, mapped injectivity to a named test, and still shipped a non-injective derive_path with two collisions reproduced on the built package. The run also turned up defects in supskill itself, including a version check that confuses the checkout with the install, and an append-only Gate 3 writeback that leaves the backlog disagreeing with its own summary table. Alpha still, and playset has never produced an MP3.
Update 2 of the supskill build log. supskill is a sprint conductor for Claude Code, built on one finding: a delegated agent has no channel to escalate a decision to a human. This update covers E7, the packaging milestone: supskill's first marketplace publish under the immutable slug supskill@supskill, its two cross-marketplace dependencies (superpowers and pm-execution) declared and resolving at install time, and a should-trigger/should-not-trigger eval loop composed from the skill-creator plugin's own evaluation harness rather than a bespoke scorer, measuring whether the conductor's SKILL.md description fires on the right prompts and stays silent on the rest. The offline suite is up to 314 passing tests, still under a second. It also covers the smaller, less glamorous discipline this sprint needed: what it took to finalize the README without accidentally taking credit for someone else's uncommitted work.
Update 1 of the supskill build log. supskill is a sprint conductor for Claude Code: it drives one sprint from a markdown backlog to a merged branch, drawing a fresh context boundary at every stage and holding three human gates, with all authoritative state on disk so it survives its own context death. The whole project rests on one finding: a delegated agent cannot ask you anything. Before writing this, I re-ran that finding on Claude Code 2.1.210, and the runtime had moved: AskUserQuestion is now unavailable inside a subagent and absent from a headless session's 62-tool manifest, and calling it returns a loud error instead of the silently-invented empty answer a cited issue reported. That makes supskill's core invariant stronger, not weaker: a delegated agent still has no channel to escalate, so nothing forces it to stop-and-ask rather than guess. This update shows the probe, the turn, and the on-disk state spine (280 offline tests) that makes a skipped gate structurally impossible.