Skip to content

REVIEW.md

Latest commit

 

History

History
72 lines (64 loc) · 3.91 KB

REVIEW.md

File metadata and controls

72 lines (64 loc) · 3.91 KB
Raw

Maintainer run-through record

2026-10-11 — authored from official documentation (mechanical verification only)

  • Course authored against the VS Code GitHub Copilot documentation at code.visualstudio.com/docs/copilot, fetched 2026-10-11 (pages: local agents and session controls, custom instructions, review and revert agent changes, Agent Skills), and the shared report-tool core. TASK.md, tests/, scenario/, starter/, solution/ are byte-identical to the Claude Code / Codex / Cursor course cores — enforced by the core-group check in tools/verify_courses.py.
  • Version reference: VS Code 1.134 with the Copilot harness on the Agent Host, as the documentation describes it on that date. No VS Code build was installed for this record.
  • Documented behaviors the lessons depend on (sourced, not observed):
    • Chat input controls: Session Target (Copilot), Agent role (Agent / Plan / Ask), Permissions (Manual permissions / Allow all), Code isolation (Folder / New Worktree; worktree sessions are always Allow all; the isolation choice exists only in the Agents window).
    • In Agent Host sessions edits are saved directly with no pending Keep/Undo state; review happens in the diff view and Source Control; Restore Checkpoint restores files and chat history but does not reverse completed terminal commands.
    • chat.checkpoints.showFileChanges and the chat.tools.edits.autoApprove glob setting for sensitive-file approval.
    • .github/copilot-instructions.md / AGENTS.md as always-on project instructions; .github/instructions/*.instructions.md attached by applyTo when the agent creates or modifies a matching file; References in the response show which instructions were used.
    • Agent Skills discovered from .github/skills/<name>/SKILL.md, where <name> must equal the skill's name field. Lesson 1 therefore installs the live FlyPython Skill (name flypython-challenges) at .github/skills/flypython-challenges/SKILL.md.
  • Commands and results (Linux, Python 3.12):
    • python verify.py starter --expect-failure — exit 0; the seven expected failures reproduced.
    • python verify.py solution — exit 0; 9/9 tests pass.
    • python verify.py — exit 1 on the shipped starter; l03/l04 [open], l01/l02/l05 [pending].
    • python tools/verify_courses.py — contract satisfied, core identical with the other report-tool members.
  • Not done: any run inside VS Code with Copilot. No Copilot account was used; every Copilot behavior described in the lessons is sourced from official documentation, not observed.

Pending before this course can be called taught-in-tool

  • One full agent-taught run-through with GitHub Copilot in VS Code ("start lesson 1" through the Lesson 5 checkpoint) on a current build, recording observed deviations from COURSE.md here — in particular whether Manual permissions prompts before the Skill URL fetch, and whether the applyTo file shows up in References as Lesson 2 expects.
  • Whether the Copilot Free plan's monthly chat request allowance lasts the whole course is unverified.

2026-11-04 — claim-instruction accuracy fix (mechanical re-verify)

  • Lesson and hint text corrected to match the shipped verifier: the default python verify.py prints no claim code for reflection checkpoints (l01/l02/l05); the code appears only after the explicit --attest ID flag for a completed reflection. L02/L05 checkpoints now name the command; L01 hint 3 no longer promises a code from the default run; evidence sections keep both command transcripts.
  • Verification: clean-copy runs of verify.py check --json with partial and full --attest sets reproduced the documented states; repository gates (make check) green at this change. No live GitHub Copilot agent run — the pending live run-through below still stands.

Deviation log

(none yet — no live run has happened)