roborev-dev roborev-lookahead-review-branch
Request a time-series look-ahead review for all commits on the current branch and present the results. A look-ahead review checks whether the change uses information that would not yet be available at the point in time it represents — also called peekahead, future leakage, or temporal leakage.
Usage
/roborev-lookahead-review-branch [--base <branch>] [--panel <name>|none]
Explicit invocation only
Invocation must be explicit: literal personal /roborev-lookahead-review-branch, or structured
Claude Code skill selection.
Requests such as “check this branch for future leakage” without one of these
explicit mechanisms must use native behavior and must not run roborev.
Sandbox access
roborev uses a local daemon. If a command fails with permission denied, the sandbox may be
blocking access to its loopback port or Unix socket. Retry the same Bash command with
dangerouslyDisableSandbox: true. Do not start or restart the daemon because a sandboxed
status probe cannot reach it.
When NOT to invoke this skill
Do NOT invoke this skill when the user is presenting or pasting existing review results. Messages that contain review findings, verdicts, or summaries are outputs — not requests to start a new review.
IMPORTANT
This skill requires you to execute bash commands to validate inputs and launch the review. The task is not complete until the background review finishes and you present the results to the user.
These instructions are guidelines, not a rigid script. Use the conversation context. Skip steps that are already satisfied. Defer to project-level CLAUDE.md instructions when they conflict with these steps.
Instructions
When the user invokes /roborev-lookahead-review-branch [--base <branch>] [--panel <name>|none]:
1. Validate inputs
If a base branch is provided, use the base-branch command snippet below; it stores and validates the ref before invoking roborev review.
If validation fails, inform the user the ref is invalid. Do not proceed.
2. Build the command
Construct the review command:
If no base branch is specified, run:
roborev review --branch --wait --type lookahead [--panel <name>|none]
If a base branch is specified, run:
read -r branch <<'ROBOREV_REF'
<branch>
ROBOREV_REF
git rev-parse --verify -- "$branch" || exit 1
roborev review --branch --wait --type lookahead --base "$branch" [--panel <name>|none]
- If
--baseis specified, include it (otherwise auto-detects the base branch) - If
--panel <name>is specified, include it (fans out to the named config panel);--panel noneforces a single-agent review
3. Run the review in the background
Launch a background task that runs the command. This lets the user continue working while the review runs.
Use the Task tool with run_in_background: true and subagent_type: "Bash":
If no base branch is specified, run:
roborev review --branch --wait --type lookahead [--panel <name>|none]
If a base branch is specified, run:
read -r branch <<'ROBOREV_REF'
<branch>
ROBOREV_REF
git rev-parse --verify -- "$branch" || exit 1
roborev review --branch --wait --type lookahead --base "$branch" [--panel <name>|none]
Tell the user that the look-ahead review has been submitted and they can continue working. You will present the results when the review completes.
4. Present the results
When the background task completes, read the output.
If the command output contains an error (e.g., daemon not running, repo not initialized, review errored), report it to the user. Suggest roborev status to check the daemon, roborev init if the repo is not initialized, or re-running the review.
Otherwise, present the review to the user:
- Show the verdict prominently (Pass or Fail)
- If there are findings, list them grouped by severity with file paths and line numbers so the user can navigate directly
- If the review passed, a brief confirmation is sufficient
Panels (multi-reviewer reviews)
If you pass --panel <name>, or a default_panel is configured for explicit
reviews, the review fans out to a panel of reviewers. In that case the
Enqueued job <id> is the synthesis (parent) job that aggregates them, and
its verdict and findings are the synthesized result across the whole panel.
Present that synthesized verdict/findings, and offer fix on that parent id —
never an individual reviewer. roborev show prints a one-line reviewers summary
(e.g. 3 reviewers: bug P, security F) for a synthesis job. --panel none
forces a single-agent review, and automatic post-commit hook reviews stay
single-agent regardless of default_panel.
5. Offer next steps
If the review has findings (verdict is Fail), offer to address them:
- "Would you like me to fix these findings? You can run
/roborev-fix <job_id>"
Extract the job ID from the review output to include in the suggestion. Look for it in the Enqueued job <id> for ... line or in the review header. For a panel review this id is the synthesis parent.
If the review passed, confirm the result and do not offer /roborev-fix.
Examples
Default branch look-ahead review:
User: /roborev-lookahead-review-branch
Agent:
- Launches background task:
roborev review --branch --wait --type lookahead - Tells user: "Look-ahead review submitted for branch. I'll present the results when it completes."
- When complete, presents the verdict and findings grouped by severity
- If findings exist: "Would you like me to address these findings? Run
/roborev-fix 1042" - If passed: "Branch look-ahead review passed with no findings."
Look-ahead review against a specific base:
User: /roborev-lookahead-review-branch --base develop
Agent:
- Validates
developresolves to a valid ref - Launches background task:
roborev review --branch --wait --type lookahead --base develop - Tells user: "Look-ahead review submitted for branch (against develop). I'll present the results when it completes."
- When complete, presents the verdict and findings
- If findings exist: "Would you like me to address these findings? Run
/roborev-fix 1043"
See also
/roborev-review-branch --type lookahead— equivalent, with additional--typeflexibility/roborev-lookahead-review— look-ahead review a single commit/roborev-fix— fix a review's findings in code