AI Dev
instructor

Landing-page workshop recovery

Recover common workshop failures without hiding them, destroying learner work, or consuming the whole class clock.

Recovery keeps the learner in the method. It does not promise that every machine or account problem can be solved during the session.

Before class

Prepare and test:

  • a public-safe source pack;
  • an approved neutral knowledge-base.md and design.md;
  • a minimal one-page plan;
  • a running Astro shell commit;
  • a content-complete local commit;
  • a reviewed deployable commit;
  • a Vercel deployment from the teaching network.

Record exact commits. Do not include credentials or private participant data.

Failure matrix

FailureDiagnose firstIn-class recoveryPreserve for follow-up
Missing source/rightsIs the primary CTA or claim blocked?Use approved instructor source pack or pair as observerLearner's unanswered content/rights questions
Unsupported Node/npmVersions and Astro's current requirementPair/use prepared environment; do not random-upgrade system toolsExact versions and official setup route
Install failsFirst error, folder, network, runtime, diskUse prepared Astro shell after explaining provenanceError output and environment
Port is busyDev-server message and printed URLUse the next offered port; open the printed URLProcess/port investigation after class
Agent cannot see filesWorkspace root and permissionsReopen exact project folder; pair if unresolvedTool/account permission issue
Astro scaffold wants overwriteCurrent files and command targetCancel; use reviewed subfolder or prepared shellPlan for moving/merging approved files
Git identity/auth failsLocal identity versus GitHub authenticationKeep local commits; deploy handoff after classCommit hash and correct account route
Production build failsFirst build error, not browser appearanceStop deploy; use last passing milestone only if it matches learner sourceBuild log and commit
Vercel cannot find appRoot folder and package.json scriptsCorrect only known root; otherwise preserve GitHub finish lineDeployment settings screenshot/log
Vercel build differsRuntime, lockfile, env, path case, commitRecord deploy blocker; do not change multiple settingsLocal clean-build evidence and remote log
Missing asset in productionFilename case/path and tracked fileCorrect the specific path/file, rebuild, recommitAsset provenance and QA rerun
Content/design behind timeCheckpoint evidenceCut sections, not review; use one page/CTADeferred list

Recovery protocol

  1. Freeze the learner's current state with git status and the last commit.
  2. State the first observed failure and what should have happened.
  3. Choose one documented recovery route.
  4. Explain what material comes from the learner and what comes from the instructor recovery project.
  5. Re-run the failed check.
  6. Commit only when the recovered state is understood.
  7. Record follow-up; return the room to the clock.

Do not use in class

  • git reset --hard or broad deletion as a first response;
  • copying a finished project without explaining its source and diff;
  • inventing participant business content;
  • putting credentials in chat or source;
  • applying multiple “maybe” fixes before rerunning the first failed check;
  • deploying an unreviewed fallback under the learner's identity.

Escalation rule

After one documented recovery attempt, move the learner to pair/observer mode if the issue involves account access, system runtime, network policy, file permission, or an unexplained destructive risk. Preserve evidence for a follow-up session.