Skip to content

rse/ase / plugin/skills/ase-repo-review

ase-repo-review

Verified

Perform a holistic, human-reviewer-style critique of the currently staged Git changes (or, with `--unstaged`, of the unstaged and untracked changes, with `--branch`, of a branch against its base, or, with `--commit`, of a single commit) and emit an approve/reject verdict with prioritized, severity-tagged, line-cited findings. Use when the user wants the staged diff, a branch, or a commit "reviewed", "critiqued", or "code-reviewed".

@${CLAUDE_SKILL_DIR}/../../meta/ase-control.md @${CLAUDE_SKILL_DIR}/../../meta/ase-skill.md @${CLAUDE_SKILL_DIR}/../../meta/ase-getopt.md

$ARGUMENTS

Procedure

    1. Set to the uncommitted changes to review: to staged and unstaged if both and are equal true, to staged if only is equal true, to unstaged if only is equal true, and to empty otherwise. Here, the unstaged changes always include the untracked files.

      1. Run the command git rev-parse --verify --quiet "<getopt-option-branch/>^{commit}". If it fails, only output the following and then STOP immediately:

      2. Determine the candidate base branches by running the command git for-each-ref --format="%(refname:short)" refs/heads/ (taken exactly as given), excluding itself. For each , run the command git rev-list --count "<candidate/>..<getopt-option-branch/>" to count the commits of the branch since its fork point from , and drop every candidate with a count of 0, as it already contains the branch. Set to the remaining candidate with the smallest count (the nearest fork point); on a tie, prefer the checked-out branch (the output of git branch --show-current), and then the first candidate. If no candidate remains, only output the following and then STOP immediately:

      3. Run the command git merge-base "<base/>" "<getopt-option-branch/>" and capture its output into .

      4. Only output the following :

    2. Determine whether there are changes at all by running the corresponding command(s) (taken exactly as given) and capturing their combined output - the bare list of changed file names - into . This is a lightweight gate; the full diff is fetched by the sub-agent in STEP 2, so capturing only the file-name list here is sufficient:

      If is not empty, set branch since base commit plus the changes in worktree and run:

      • If is equal staged and unstaged:

        git -C "<worktree-dir/>" diff --name-only "<base-commit/>"

        git -C "<worktree-dir/>" ls-files --others --exclude-standard

      • If is equal staged:

        git -C "<worktree-dir/>" diff --cached --name-only "<base-commit/>"

      • If is equal unstaged:

        git diff --name-only "<base-commit/>..<getopt-option-branch/>"

        git -C "<worktree-dir/>" diff --name-only

        git -C "<worktree-dir/>" ls-files --others --exclude-standard

      From the output of the ls-files command, skip every listed entry below the .ase/ directory -- it carries ASE's own state, and hence is never part of the user's change set.

      Otherwise, set branch since base commit and run:

      git diff --name-only "<base-commit/>..<getopt-option-branch/>" Set branch since base commit and run:

      git diff --name-only "<base-commit/>..<getopt-option-branch/>" Set commit and run:

      git show --format= --name-only --diff-merges=first-parent "<commit-id/>" Set staged and unstaged and run:

      git diff --name-only HEAD

      git ls-files --others --exclude-standard

      From the output of the second command, skip every listed entry below the .ase/ directory -- it carries ASE's own state, and hence is never part of the user's change set. Set unstaged and run:

      git diff --name-only

      git ls-files --others --exclude-standard

      From the output of the second command, skip every listed entry below the .ase/ directory -- it carries ASE's own state, and hence is never part of the user's change set. Set staged and run:

      git diff --cached --name-only HEAD

  1. The project source artifacts are classified as a black box, so the user does not want the changes scrutinized or their findings surfaced (an explicitly given severity floor lifts this skip). Skip the entire review investigation: do not invoke the Agent tool and do not read any change, set to the empty list, set to 0, set to SKIPPED (boxing: black), set to a one-line neutral restatement of the change intent derived solely from STEP 1, and proceed directly to STEP 3.

    First, use the following to give a hint on this step:

    Dispatch the review investigation to a sub-agent via the Agent tool so that no investigation details leak into the user-visible transcript. The sub-agent performs the silent reading, the read-only repository probing, and the critique; only its final structured return value is consumed here.

    Set thorough if is equal true, and focused otherwise.

    For this, invoke exactly once the tool:

        Agent(
            description:       "Review Investigation",
            subagent_type:     "ase:ase-repo-review",
            prompt:            "Review in <mode/> mode the changes of scope: <scope/>",
            run_in_background: false
        )
    

    Parse the single result message of the Agent tool as a JSON object, set to its summary field (a single crisp sentence reconstructing the change's intent), and set to its findings field (a list).

    Then derive the overall from : set to REJECT - DEMANDS CHANGES if any finding in has a severity field of HIGH; otherwise set to APPROVE. The verdict is derived before the severity floor below, so the floor only affects which findings are rendered, never the verdict.

    Then determine the effective severity floor and its origin : define the ordinal rank LOW=1, MEDIUM=2, HIGH=3.

    Then apply the effective severity floor : Keep a finding in if and only if its severity field is ACCEPTED or rank(severity) is greater than or equal to rank(<floor/>); drop all other findings, but set to the number of dropped findings. With the floor LOW, all findings are kept. ACCEPTED findings are never dropped. The floor affects only which findings are rendered, never the derived above.

    You MUST NOT output anything else in this STEP 2.

    1. Use the following to output the overall review in and the reconstructed intent :

      REVIEW VERDICT:

      CHANGE INTENT:

      You MUST NOT output anything else in this STEP 3.1.

    2. NO FINDINGS: , nothing to flag.

    3. Then render a three-column table with one row per finding by using the following output . For each finding, repeat the third line, set to its severity field, set to its dimension field, set to its location field, and set to its finding field.

      In the column, mark up the file:line reference as code (with backticks) and prepend it with ▢ - keep the sub-agent's own :N / :N-M line citation intact and do not append any further line-count decoration.

      Because the text is free-form Markdown, before emitting any row you MUST escape every literal | pipe character inside and as \| so it cannot break the table column structure.

      Keep the overall report concise and brief. Do not output any further explanation.

    4. HIDDEN FINDINGS: finding(s) below severity suppressed ().

    5. Finally, give the closing hints by expanding the following (which, depending on the configured , may expand into nothing and hence emit no output at all):

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 implementationStars: 322kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 322kUpdated Claude CodeCodexVerifiedInstall
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 fallbackStars: 322kUpdated Claude CodeCodexVerifiedInstall
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 alwaysStars: 322kUpdated Claude CodeCodexVerifiedInstall