review-labels #14
4
.github/pull_request_template.md
vendored
@@ -9,7 +9,5 @@
|
||||
|
||||
## Checklist
|
||||
<!-- Complete after creating PR -->
|
||||
- [ ] Linked issues
|
||||
- [ ] Reviewed changes
|
||||
- [ ] Reviewed changes (or use `Review/AI` label)
|
||||
- [ ] Updated comments/documentation
|
||||
|
||||
2
.github/workflows/code-review.yml
vendored
@@ -2,7 +2,7 @@ name: Code review
|
||||
|
||||
on:
|
||||
pull_request:
|
||||
types: [opened, synchronize, reopened]
|
||||
types: [opened, synchronize, reopened, labeled]
|
||||
|
||||
jobs:
|
||||
review:
|
||||
|
||||
1
.github/workflows/ticket-refinement.yml
vendored
@@ -3,6 +3,7 @@ name: Ticket refinement
|
||||
on:
|
||||
issues:
|
||||
types: [labeled]
|
||||
|
||||
jobs:
|
||||
format:
|
||||
runs-on: ubuntu-latest
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
{
|
||||
"name": "@ztimson/ai-agents",
|
||||
"version": "0.1.1",
|
||||
"version": "0.1.2",
|
||||
"description": "AI agents",
|
||||
"keywords": ["ai", "review"],
|
||||
"author": "ztimson",
|
||||
|
||||
@@ -181,8 +181,9 @@ Output ONLY markdown. No explanations, labels, or extra formatting.`});
|
||||
const hasDuplicates = (await ai.language.ask(`ID: ${issueData.id}\nTitle: ${title}\n\`\`\`markdown\n${body}\n\`\`\``, {
|
||||
system: `Your job is to identify duplicates. Respond ONLY with the duplicate's ID number or "NONE" if no match exists\n\n${dupes}`
|
||||
}))?.pop()?.content;
|
||||
|
||||
// Handle duplicates
|
||||
if(hasDuplicates && hasDuplicates !== 'NONE' && (dupeId = dupeIds.find(id => id == hasDuplicates.trim()))) {
|
||||
if(hasDuplicates && !hasDuplicates.toUpperCase().includes('NONE') && (dupeId = dupeIds.find(id => id == hasDuplicates.trim())) != null && dupeId != issueData.id) {
|
||||
|
ztimson marked this conversation as resolved
|
||||
await fetch(`${git}/api/v1/repos/${owner}/${repo}/issues/${ticket}/comments`, {
|
||||
method: 'POST',
|
||||
headers: {'Authorization': `token ${auth}`, 'Content-Type': 'application/json'},
|
||||
|
||||
@@ -19,12 +19,20 @@ dotenv.config({path: '.env.local', override: true, quiet: true, debug: false});
|
||||
owner = process.env['GIT_OWNER'],
|
||||
repo = process.env['GIT_REPO'],
|
||||
auth = process.env['GIT_TOKEN'],
|
||||
labelEnabled = process.env['LABEL_ENABLED'] || 'Review/AI',
|
||||
pr = process.env['PULL_REQUEST'],
|
||||
host = process.env['AI_HOST'],
|
||||
model = process.env['AI_MODEL'],
|
||||
token = process.env['AI_TOKEN'];
|
||||
|
||||
console.log(`Reviewing: ${root}\n`);
|
||||
const info = await fetch(`${git}/api/v1/repos/${owner}/${repo}/pulls/${pr}`)
|
||||
|
ztimson marked this conversation as resolved
assistant
commented
Missing error handling: The fetch call should handle errors. If the API call fails, Missing error handling: The fetch call should handle errors. If the API call fails, `info` will be undefined and line 30 will throw an error when trying to access `info.labels`.
assistant
commented
Missing error handling: The fetch call should handle network errors with a .catch() handler. If the API call fails due to network issues, the promise will reject and crash the script before reaching line 30. Missing error handling: The fetch call should handle network errors with a .catch() handler. If the API call fails due to network issues, the promise will reject and crash the script before reaching line 30.
|
||||
.then(async resp => { return resp.ok ? resp.json() : throw new Error(`${resp.status} ${await resp.text()}`); });
|
||||
|
ztimson marked this conversation as resolved
Outdated
assistant
commented
Logic error: The condition Logic error: The condition `info.labels?.length > 0 || !info.labels.some(l => l.name === labelEnabled)` will always be true when labels exist. This should use `&&` instead of `||`. Currently, if labels.length > 0, it skips even if the enabled label is present. The correct logic should be: `!info.labels?.some(l => l.name === labelEnabled)` or `info.labels?.length === 0 || !info.labels.some(l => l.name === labelEnabled)`
assistant
commented
Syntax error: Syntax error: `throw new Error(...)` cannot be used directly in a ternary expression. This should be wrapped in a block: `resp.ok ? resp.json() : (() => { throw new Error(...); })()` or use an if-else statement instead.
|
||||
if(info.labels?.length > 0 || !info.labels.some(l => l.name === labelEnabled)) {
|
||||
|
ztimson marked this conversation as resolved
Outdated
assistant
commented
Logic error: The condition Logic error: The condition `info.labels?.length > 0 && !info.labels.some(l => l.name === labelEnabled)` will skip the review when labels exist but the enabled label is NOT present. This appears backwards - it should run when the label IS present. Consider: `if(!info.labels?.some(l => l.name === labelEnabled))` to skip when the label is missing.
|
||||
console.log('Skipping');
|
||||
return process.exit();
|
||||
}
|
||||
|
||||
const branch = process.env['GIT_BRANCH'] || await $`cd ${root} && git symbolic-ref refs/remotes/origin/HEAD`;
|
||||
const comments = [];
|
||||
const commit = await $`cd ${root} && git log -1 --pretty=format:%H`;
|
||||
@@ -53,7 +61,7 @@ dotenv.config({path: '.env.local', override: true, quiet: true, debug: false});
|
||||
...options,
|
||||
model: [host, model],
|
||||
path: process.env['path'] || os.tmpdir(),
|
||||
system: `You are a code reviewer. Analyze the git diff and use the \`recommend\` tool for EACH issue you find. You must call \`recommend\` exactly once for every bug or improvement opportunity directly related to changes. Ignore formatting recommendations. After making all recommendations, provide some concluding remarks about the overall state of the changes.${existingComments}`,
|
||||
system: `You are a code reviewer. Analyze the git diff and use the \`recommend\` tool for EACH issue you find. You must call \`recommend\` exactly once for every bug or improvement opportunity directly related to changes. Ignore formatting recommendations. After making all recommendations, provide a quick 75 words or less sitrep.${existingComments}`,
|
||||
|
ztimson marked this conversation as resolved
Outdated
assistant
commented
Typo in system prompt: 'sitrep' is informal military jargon. Consider using more professional language like 'summary' or 'overview' for better clarity in a code review context. Typo in system prompt: 'sitrep' is informal military jargon. Consider using more professional language like 'summary' or 'overview' for better clarity in a code review context.
assistant
commented
Typo in system prompt: 'sitrep' is informal military jargon. Consider using more professional language like 'summary' or 'overview' for better clarity in a code review context. Typo in system prompt: 'sitrep' is informal military jargon. Consider using more professional language like 'summary' or 'overview' for better clarity in a code review context.
|
||||
tools: [{
|
||||
name: 'read_file',
|
||||
description: 'Read contents of a file',
|
||||
|
||||
Improved duplicate detection logic, but the condition is overly complex. Consider simplifying:
dupeId = dupeIds.find(id => id == hasDuplicates.trim()); if(hasDuplicates && !hasDuplicates.toUpperCase().includes('NONE') && dupeId != null && dupeId != issueData.id)could be clearer with early returns or better variable naming.Potential type coercion bug: Using
==instead of===indupeIds.find(id => id == hasDuplicates.trim())anddupeId != issueData.idmay cause unexpected behavior if IDs are sometimes strings and sometimes numbers. Use strict equality (===and!==) for more predictable comparisons.Potential type coercion bug: Using
==instead of===indupeIds.find(id => id == hasDuplicates.trim())may cause unexpected behavior if IDs are sometimes strings and sometimes numbers. Use strict equality (===) for more predictable comparisons.Redundant check: The condition
dupeId != issueData.idis unnecessary since the search is limited to 3 results and unlikely to return the current issue. However, if kept, use strict inequality (!==) instead of!=for consistency.