BM-2750: Improvements in Agent Workflow - #1819
Conversation
github-actions
Bot
changed the title
capossele
left a comment
capossele
left a comment
There was a problem hiding this comment.
Left a few comments but good overall
2694d99 to
bcfcf1d
Compare
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
Bugbot Autofix is kicking off a free cloud agent to fix these issues. This run is complimentary, but you can enable autofix for all future PRs in the Cursor dashboard.
| [[ -n "$TUNNEL_PID" ]] && kill "$TUNNEL_PID" 2>/dev/null && wait "$TUNNEL_PID" 2>/dev/null | ||
| [[ -n "$SERVE_DIR" ]] && rm -rf "$SERVE_DIR" | ||
| exit $exit_code | ||
| } |
There was a problem hiding this comment.
Force-quit exits with success code instead of failure
Medium Severity
The cleanup function captures $? at its start and calls exit $exit_code. When handle_sigint calls cleanup() on the 3rd Ctrl-C, $? is 0 from the preceding printf, so the script exits with code 0 (success) instead of 1 (failure). The exit 1 after the cleanup call in handle_sigint is dead code because cleanup already exits the script.
Additional Locations (1)
| } | ||
| } | ||
| None | ||
| } |
There was a problem hiding this comment.
YAML frontmatter parser truncates multiline description values
Low Severity
parse_description only captures text on the same line as description:. If a SKILL.md uses standard YAML multiline syntax (folded >, literal |, or flow continuation), only the first line is returned and the rest is silently dropped. This affects both local discovery and GitHub-fetched skills, producing truncated descriptions in --list output and interactive selection.
Relocate boundless-overview and first-request skills to the standard Claude Code skills directory so they are discoverable without embedding.
This skill is no longer needed now that skills are discovered dynamically rather than compiled into the binary.
The dev subcommand for symlinking skill sources is no longer needed since skills are now discovered dynamically from .claude/skills/ or GitHub.
Replace compile-time embedded skill files with runtime discovery from: - The local repo's .claude/skills/ directory (when inside the Boundless repo) - GitHub boundless-xyz/boundless main branch (when outside the repo) Update install tests to reflect the new dynamic discovery behavior.
bcfcf1d to
2bb9bc9
Compare
|
You have used all of your free Bugbot PR reviews. To receive reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial. |
New Claude Code skill that uses Cloudflare Quick Tunnels to self-host guest program binaries and input files, eliminating the need for Pinata, S3, or any third-party storage account when submitting proof requests. Includes: - scripts/self-host.sh: long-running script that starts a local HTTP server + cloudflared tunnel, builds request YAML, submits via CLI, and polls for fulfillment. 3x Ctrl-C guard prevents accidental tunnel death while provers are downloading files. - scripts/check-prerequisites.sh: validates cloudflared, python3, curl, boundless CLI, and cast are installed - SKILL.md: agent-facing guide covering wallet setup, CLI config, guest program building, and the full serve/submit/wait workflow - PLAN.md: design doc with architecture and open questions
Address review comments from @capossele: - Use GITHUB_TOKEN env var as Bearer token in GitHub API requests if available (unauthenticated: 60 req/hr, authenticated: 5,000 req/hr) - Detect 403 + x-ratelimit-remaining: 0 and show a helpful error message pointing users to create a token
The skill install CLI is being replaced by a simpler agent-friendly approach using llms.txt and direct skill discovery. Remove the skill subcommand, its install logic, and all related tests. Also fix self-host.sh force-quit to exit with 130 (SIGINT convention) instead of 0.
2bb9bc9 to
4afdacd
Compare
|
You have used all of your free Bugbot PR reviews. To receive reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial. |
- Merge first-request + self-hosted-storage → requesting (no Pinata, self-host default) - Rename boundless-overview → boundless-cli (focused CLI skill with cross-references) - Rename deploy-prover → setup-prover (fix lowercase skill.md → SKILL.md) - Prefix internal ops skills: ops-infra-deploy, ops-add-new-chain, ops-telemetry-query - Add 'Internal — for Boundless team members only' to ops skill descriptions - Update boundless-cli env vars table from source (added 8 missing vars)
|
You have used all of your free Bugbot PR reviews. To receive reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial. |
|
You have used all of your free Bugbot PR reviews. To receive reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial. |
|
You have used all of your free Bugbot PR reviews. To receive reviews on all of your PRs, visit the Cursor dashboard to activate Pro and start your 14-day free trial. |
2 participants
Changes
Skills refactoring
crates/boundless-cli/skill/to.claude/skills/as source of truthcontributing-skillsskillboundless skillsubcommand entirely, in favor of skill discovery via the docsSkill consolidation & renaming
first-request+self-hosted-storage→requesting— single skill with two paths (self-host via Cloudflare tunnel or replay a recent request). No Pinata/S3 dependency.boundless-overview→boundless-cli— focused CLI reference with cross-references to other skills, updated env vars from sourcedeploy-prover→setup-prover— also fixed lowercaseskill.md→SKILL.mdops-(ops-infra-deploy,ops-add-new-chain,ops-telemetry-query) and mark descriptions as internal-onlyFinal skill layout
boundless-clirequestingsetup-proverops-infra-deployops-add-new-chainops-telemetry-query