Skip to content

fix(ce-demo-reel): prevent secrets in recorded demos#664

Merged
tmchow merged 1 commit intomainfrom
tmchow/demo-reel-no-secrets
Apr 24, 2026
Merged

fix(ce-demo-reel): prevent secrets in recorded demos#664
tmchow merged 1 commit intomainfrom
tmchow/demo-reel-no-secrets

Conversation

@tmchow
Copy link
Copy Markdown
Collaborator

@tmchow tmchow commented Apr 24, 2026

Summary

Recorded demos must never contain credentials. A new "Never Record Secrets" ground rule in ce-demo-reel codifies that across all capture tiers: plan secrets out of frame, never substitute placeholders inline, and scan before upload.

The prior skill had only a single post-capture bullet in one tier, and didn't prevent the subtler failure mode — a visible export API_KEY=REDACTED "mask" overwrites the real env var, producing a broken demo (401, 0 credits remaining, empty output) that still leaks the variable name. The rule is built around one principle: secrets should affect the environment, not the visible transcript.

What's new

  • Ground rule in SKILL.md — three directives (plan out of frame, don't substitute placeholders, scan before upload) with the runtime-state failure called out explicitly, plus the positive principle "show the authenticated result, not the auth step."
  • Concrete VHS pattern in tier-terminal-recording.md — clean HOME, secrets in Hide, clear, then Show. Full skeleton included as the teaching artifact.
  • Tier-specific guards for browser reels (no DevTools, clean URL bar, avoid raw-credential pages), screenshot reels (authored-content discipline — no real or fake-looking credentials), and static screenshots (both paths).
  • Drift check — a broken visible command (401 Unauthorized, 0 credits remaining, empty output where data was expected) usually means a visible export SECRET=... after Show overwrote the real env. Fix the .tape and re-record rather than ship.

Compound Engineering
Claude Code

Introduce "Never Record Secrets" as a ground rule: set credentials before
recording, never substitute placeholders inline, scan artifacts before
upload. Add the concrete VHS Hide/clear/Show pattern plus per-tier
guidance for browser, screenshot, and static captures, and a drift
check for broken demos caused by re-exporting a fake value after Show.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@tmchow tmchow changed the title feat(ce-demo-reel): prevent secrets in recorded demos fix(ce-demo-reel): prevent secrets in recorded demos Apr 24, 2026
Copy link
Copy Markdown

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 00271cf3a7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

tmchow added a commit that referenced this pull request Apr 24, 2026
Codex review flagged that the VHS tape example embedded a literal
API key inside `export API_KEY='real-secret-value'`. Even with Hide
keeping it out of the GIF, the .tape itself is a persistent file
that can leak through temp-dir retention, debug copies, or sharing.

Switch the pattern: the invoking shell carries the secret (direnv,
source .env, or export before vhs), VHS inherits it, and the tape
file never contains the literal. Planning bullet, example comment,
"Why this shape" note, and drift-check remediation all follow.

Resolves: #664 (comment)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Copy link
Copy Markdown

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 78a82c6cd1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@tmchow tmchow force-pushed the tmchow/demo-reel-no-secrets branch from 78a82c6 to 00271cf Compare April 24, 2026 04:38
@tmchow tmchow merged commit 9ddcd22 into main Apr 24, 2026
4 checks passed
@github-actions github-actions Bot mentioned this pull request Apr 24, 2026
@tmchow tmchow mentioned this pull request Apr 24, 2026
3 tasks
tmchow added a commit that referenced this pull request Apr 24, 2026
PR #663 squash-merged as `feat(ce-commit-push-pr):` — adding two
short-circuits that skip a blocking prompt when the agent just authored
the change. The intent was a UX/flow fix, not a new capability, so the
correct conventional prefix is `fix:`. Since the prefix drove a minor
bump, the open release PR (#661) is pinning the linked `cli` /
`compound-engineering` group at 3.1.0.

This empty commit reclassifies the pending release to 3.0.2 via
`Release-As:` footers so the patch-level fixes in this window
(#660 ce-update, #664 ce-demo-reel, #663 ce-commit-push-pr) land as
3.0.2 instead of 3.1.0. The feat prefix remains in history and may
still render under a Features heading in the generated changelog —
hand-edit the release PR body before merge if so.

Release-As: cli@3.0.2
Release-As: compound-engineering@3.0.2
@github-actions github-actions Bot mentioned this pull request Apr 24, 2026
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