Skip to content

docs(sdd): record install-research explores and the defect-6 verdict - #373

Merged
pvilarim merged 2 commits into
masterfrom
claude/record-install-research
Aug 13, 2026
Merged

docs(sdd): record install-research explores and the defect-6 verdict#373
pvilarim merged 2 commits into
masterfrom
claude/record-install-research

Conversation

@pvilarim

Copy link
Copy Markdown
Owner

Records three parallel explore sessions as durable research artifacts, so the 1.15.0 proposal session reads them from the repo instead of chat scrollback. Exploration only — no kit code changes.

Root cause all three converge on: the hub validates itself. Seven defects in 40 minutes of real consumer use; zero in 369 CI PRs.

  • explore-consumer-install-defects — seven defects from a real Windows/APP install, plus the adversarial verdict on defect 6: the pointer-rewrite plan died under two clean-context agents (78 dangling references in 23 shipped files, 10–11 with no delivered target, 10 spec lines locking the pointer text, KEEP-on-exists blocking existing installs). Decided fix: ship the guide itself (template mirror inside sdd-kit/ + one MANIFEST COPY entry + parity gate + honest delta to the two "never receives it" spec sentences — which restores the founding design''s original intent from 2026-06-17).
  • explore-agent-mediated-install — decisions D1–D7 for the single-acquisition-path future (1.16.0 scope).
  • explore-python-onboarding-ux — Python detection states, alert ordering, the flock finding, macOS portability layers, and the confirmed hypothesis that no pre-1.14.0 install could ever complete on macOS (BSD realpath rejects the GNU flags), including one retraction after the real Mac report.

🤖 Generated with Claude Code

pvilarim and others added 2 commits August 13, 2026 20:05
…rdict

Three parallel explore sessions converged on one root cause: the hub
validates itself, and nobody ever runs the installer against a repository
that is not the hub. Seven defects surfaced in 40 minutes of real use;
zero in 369 CI PRs.

- explore-consumer-install-defects: seven defects from a real Windows/APP
  install (kit-root asymmetry, cp-same-file, _template failing the gate,
  verify.sh hub misdetection x2, dangling guide pointers, project.md never
  materializing), plus the adversarial verdict on defect 6: the
  pointer-rewrite recommendation died (78 references in 23 files, 10-11
  with no delivered target, 10 spec lines locking the text, KEEP-on-exists
  blocking existing installs); the decided fix is shipping the guide
  itself via a MANIFEST COPY entry with a template mirror inside sdd-kit/.
- explore-agent-mediated-install: decisions D1-D7 (single acquisition
  path, whole kit in target, detected profile, guaranteed language
  question, imperative README, INSTALL.md direction, reinstall
  comparison), with the D1 consequence that two of the seven defects
  disappear by path deletion.
- explore-python-onboarding-ux: Python states S1-S5, alert ordering,
  session-lock finding (flock absent on both target platforms), macOS
  portability layers (bash 3.2, BSD sed/realpath), the confirmed
  hypothesis that no pre-1.14.0 install could complete on macOS, and the
  retraction of the sha256sum claim after the real Mac report.

Exploration artifacts only - no kit code changes.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@pvilarim
pvilarim merged commit 2b73ccf into master Aug 13, 2026
1 check passed
@pvilarim
pvilarim deleted the claude/record-install-research branch August 13, 2026 23:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant