Skip to content

fix: update demo output, Under the Hood claims, and demo script for CLI accuracy#72

Merged
juhgiyo merged 4 commits into
mainfrom
claude/511-audit-under-the-hood
Mar 25, 2026
Merged

fix: update demo output, Under the Hood claims, and demo script for CLI accuracy#72
juhgiyo merged 4 commits into
mainfrom
claude/511-audit-under-the-hood

Conversation

@juhgiyo
Copy link
Copy Markdown
Contributor

@juhgiyo juhgiyo commented Mar 25, 2026

Summary

Aligns the strawpot.com website with the current CLI behavior after strawpot run was not shipped (issue strawpot/strawpot#509 and PR strawpot/strawpot#512 were closed).

Output block (index.astro)

  • Replace aspirational checkmark output with format matching the real TerminalProgressRenderer from cli/src/strawpot/progress.py
  • New output shows: Session started, > Delegating to {role}..., ✓ {role} completed ({duration}), ✓ Session complete
  • Stage names use real role slugs (implementer, code-reviewer)

Under the Hood claims (index.astro)

  • Git worktree isolation: "configurable merge strategies" → "configurable branch cleanup" (no merge strategy config exists; cleanup_branches controls branch deletion)
  • Structured tracing: "Full reconstruction" → "Audit any session" (no replay feature exists; JSONL logs support auditing)
  • Four other claims verified accurate — left untouched

Demo script (demo-script.yaml)

  • Add demo-script.yaml to the repo (was previously untracked)
  • Use strawpot start --task instead of strawpot run (the run alias was never added)

Closes strawpot/strawpot#510
Closes strawpot/strawpot#511

Test plan

  • Output block matches TerminalProgressRenderer format
  • Under the Hood claims audited against source code
  • Demo script uses strawpot start --task (not strawpot run)
  • No remaining references to strawpot run in tracked files
  • Visual: npm run dev and verify both sections render correctly

🤖 Generated with Claude Code

Audit findings:
- Recursive delegation: accurate (policy section in config exists) — no change
- Git worktree isolation: "configurable merge strategies" not implemented,
  changed to "configurable branch cleanup" (cleanup_branches config exists)
- Persistent memory: accurate — no change
- Any agent runtime: accurate — no change
- Roles as Markdown: accurate — no change
- Structured tracing: content-addressed artifacts exist in trace.py,
  but "Full reconstruction" is not implemented — changed to "Audit any session"

Closes strawpot/strawpot#511

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace aspirational checkmark output with format that matches the
actual TerminalProgressRenderer from the CLI progress system: session
start, delegation stages with role names and durations, and session
complete line.

Closes strawpot/strawpot#510

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The `strawpot run` command was not added (issue #509 and PR #512 were
closed). Update demo-script.yaml to use the actual CLI command.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@juhgiyo juhgiyo changed the title fix: audit and soften Under the hood claims for accuracy fix: update demo output, Under the Hood claims, and demo script for CLI accuracy Mar 25, 2026
Absent key and explicit null are semantically identical in YAML.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@juhgiyo juhgiyo merged commit 3c3f3f0 into main Mar 25, 2026
1 check passed
@juhgiyo juhgiyo deleted the claude/511-audit-under-the-hood branch March 25, 2026 15:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

1 participant