hashgraph-online/awesome-codex-plugins / plugins/beastlabai/multi-llm-plugin/skills/multi-llm
multi-llm
FlaggedOrchestrate plan reviews, task generation, implementation and code reviews across several LLM CLIs in parallel, or ask every model one free-text question about a plan. Use when the user says: review my plan, multi-llm, run a code review, generate tasks from a plan, implement this plan, ask the models about my plan.
prompts/commit_review.txt
View on GitHubYou are reviewing the code changes made by a range of git commits.
## Commit Brief
{commit_brief_path}
The commit brief stands in for a plan: it lists every commit of the range
with its full message, the diff stat, and any focus or context plan given.
Read it first.
## Commits Under Review
{commit_range_line}
{commit_list}
This list shows at most the 20 newest commits. When the range has more, the
pointer line ("… and <k> more (full list: …)") names where the rest are: the
commit brief, or a `git log` command. The range is every one of its commits,
not only the ones listed here.
{focus_block}{plan_block}## Files Changed in the Range
Where this list came from: {scope_source}
{tracked_files_list}
This list is only the head of the review scope. The full scope is the scope manifest, {scope_file}: a path it lists is in scope whether or not the list above shows it. The working tree is not what is reviewed: ignore uncommitted changes in the repo, including edits to files the manifest lists.
## Diff Access
{diff_access}
{drift_block}## What Is Reviewed
The commit messages, taken together, are the stated intent of the range.
Each message is its author's claim about the commit: check it against the
diff, and never take it as a description of what the diff does.
What you review is the net change of the whole range: the diff from the
range's base to its last commit, one diff file per changed file. The net diff
hides the intermediate states. A bug one commit added and a later commit
removed is not there to find, and a message that describes one step is
checked against what the net change leaves of it. Review the range as a
whole: a commit-by-commit review is out of scope.
## Your Task
1. Read the commit brief (see Commit Brief above): the commits, their messages and, when given, the focus and the context plan
2. For EACH file the scope manifest lists, examine its change in its diff file (under {diff_dir}). An entry with no diff file says why on its own line
3. Compare the net change against what the messages, taken together, say the range does: report what they say it does and it does not (type "missing"), and what it changes that no message accounts for (type "scope")
4. Identify any issues
Look for:
- **Bugs**: Logic errors, incorrect behavior, edge cases
- **Security issues**: Vulnerabilities, unsafe patterns, and data exposure through logs, telemetry, caches, test output or committed files
- **Regressions**: Behaviour that worked before the range and no longer does: a removed guard, a changed default, a caller or a test left behind
- **Commit interactions**: How the commits combine, e.g. a later commit that breaks an assumption an earlier one relied on, or two commits that each look right and are wrong together
- **Missing functionality**: Something a commit message says the range does that the net change does not do
- **Tests that test nothing**: Tests in the change offered as evidence for a behaviour that would still pass with that behaviour missing or broken: they never reach the code that implements it, mock out that code, assert something that cannot fail, or are skipped unconditionally or for a reason that does not apply
- **Scope creep**: Changes that no commit message accounts for
- **Quality issues**: Performance problems, code smells
{working_limits}
{scope_policy_guidance}
## Output Format
Return ONLY a valid JSON array with no additional text. Each issue must have this structure:
{{
"title": "Short issue title (5-10 words)",
"desc": "What is wrong and why it matters, with evidence where it helps",
"importance": "high",
"file": "path/to/file.py",
"line_range": [10, 15],
"type": "bug",
"anchor_text": "exact code snippet identifying the location",
"introduced_by": "e4f5a6b",
"scope_relation": "in_scope",
"scope_reason": "One short line saying why, against the scope baseline",
"scope_reference": "the commit message or context plan section it rests on, or empty"
}}
Field requirements:
- "title": string, concise summary (required)
- "desc": string, 2-4 sentences: what is wrong and why it matters, with the evidence where it helps (what in the cited lines shows it, another location as path:line, or the commit message the change contradicts). If a commit message, the context plan or the surrounding code shows the intended behaviour, say what it is. You don't need to propose a fix; whoever applies the finding decides it. But if the right fix is a design choice that neither settles, propose the one you would make, so that validation can weigh it. (required)
- "importance": exactly "high", "medium", or "low" (required)
- high = bugs, security issues, incorrect behavior
- medium = missing functionality, significant improvements
- low = style issues, minor improvements
- "file": string, relative path to affected file (required)
- "line_range": [start, end] line numbers as they are at the range's last commit (the new side of the diff), or null (required)
- "type": exactly "bug", "missing", "improvement", "style", or "scope" (required)
- "anchor_text": string, 10-50 char code snippet for locating issue (required)
- for a secret, use a snippet without the value, such as the assignment's left-hand side ("self.api_key ="). If no 10-50 char snippet leaves the value out, a shorter snippet or "" is fine; give line_range so file and line_range locate it
- "introduced_by": string or null (optional): the short SHA, as the commit list above shows it, of the commit that introduced the issue. For code a merge brought in, give the merge. Use null when you cannot tell. The plugin checks the value against the commits and git blame, so give your reading, never a guess
- "scope_relation": exactly "in_scope", "scope_extension", or "unclear" (required): how the issue's proposed action relates to the scope baseline, as the scope section above defines them
- "scope_reason": string, one short line saying why, against the scope baseline (required)
- "scope_reference": string, the baseline section or text the classification rests on, or "" (optional)
If a finding involves a secret (a key, token or password), say where it is and never copy its value into any field. That includes evidence: cite where the value is, and leave it out of any quote.
If no issues are found, return an empty array: []
## Output Location
{output_location}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |