Skip to content

Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.

Invoke Claude Code CLI directly for issue triage

Rahul Tarak6744e9993b1e
diff --git a/.github/workflows/issue-triage.yml b/.github/workflows/issue-triage.yml
index 22955589a..b66b337fc 100644
--- a/.github/workflows/issue-triage.yml
+++ b/.github/workflows/issue-triage.yml
@@ -81,54 +81,81 @@ jobs:
}
EOF
+ - name: Install Claude Code CLI
+ run: npm install -g @anthropic-ai/claude-code
+
- name: Classify issue with Claude Sonnet
id: classify
- uses: anthropics/claude-code-action@v1
env:
+ ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
ISSUE_TITLE: ${{ github.event_name == 'workflow_dispatch' && steps.issue-context.outputs.issue_title || github.event.issue.title }}
ISSUE_BODY: ${{ github.event_name == 'workflow_dispatch' && steps.issue-context.outputs.issue_body || github.event.issue.body }}
- with:
- anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
- show_full_output: true
- prompt: |
- Read the issue title and body from the ISSUE_TITLE and ISSUE_BODY environment variables.
-
- Classify the issue for routing and determine whether it is a bug report.
-
- Allowed categories:
- - docs
- - py
- - ts
- - cli
- - tool-request
- - feature-request
- - other
-
- Use semantic meaning, not only keywords.
- Choose tool-request when the issue is primarily asking for a new integration, toolkit, app, or tool.
- Choose feature-request when the issue is asking for a product, SDK, docs, or CLI improvement that is not a tool request.
- Choose cli only when the issue is specifically about the CLI experience or ts/packages/cli.
- Choose py only when the issue is primarily about the Python SDK.
- Choose ts only when the issue is primarily about the TypeScript SDK or TS packages other than the CLI.
- Choose docs only when the issue is primarily about documentation.
- Set is_bug to true when the issue is primarily reporting broken behavior, a defect, an error, a regression, or unexpected behavior.
- Set is_bug to false for feature requests, questions, maintenance, cleanup, and tool requests.
- Set should_comment to true only when the issue appears incorrect, non-actionable, or missing the information needed to reproduce or investigate.
- Set should_comment to false otherwise.
-
- If should_comment is true, provide comment_body as a short issue comment addressed to the reporter.
- The comment must:
- - clearly say it is an automated Claude triage note
- - clearly say Claude cannot reply back in a conversation
- - ask for the specific missing details needed to investigate, especially repro steps, expected behavior, actual behavior, environment, logs, or screenshots when relevant
- - stay concise and professional
-
- If should_comment is false, set comment_body to an empty string.
- claude_args: |
- --model claude-sonnet-4-6
- --json-schema /tmp/issue-triage-schema.json
- --max-turns 10
- --verbose
+ run: |
+ set -euo pipefail
+
+ PROMPT=$(cat <<'EOF'
+ Read the issue title and body from the ISSUE_TITLE and ISSUE_BODY environment variables.
+
+ Classify the issue for routing and determine whether it is a bug report.
+
+ Allowed categories:
+ - docs
+ - py
+ - ts
+ - cli
+ - tool-request
+ - feature-request
+ - other
+
+ Use semantic meaning, not only keywords.
+ Choose tool-request when the issue is primarily asking for a new integration, toolkit, app, or tool.
+ Choose feature-request when the issue is asking for a product, SDK, docs, or CLI improvement that is not a tool request.
+ Choose cli only when the issue is specifically about the CLI experience or ts/packages/cli.
+ Choose py only when the issue is primarily about the Python SDK.
+ Choose ts only when the issue is primarily about the TypeScript SDK or TS packages other than the CLI.
+ Choose docs only when the issue is primarily about documentation.
+ Set is_bug to true when the issue is primarily reporting broken behavior, a defect, an error, a regression, or unexpected behavior.
+ Set is_bug to false for feature requests, questions, maintenance, cleanup, and tool requests.
+ Set should_comment to true only when the issue appears incorrect, non-actionable, or missing the information needed to reproduce or investigate.
+ Set should_comment to false otherwise.
+
+ If should_comment is true, provide comment_body as a short issue comment addressed to the reporter.
+ The comment must:
+ - clearly say it is an automated Claude triage note
+ - clearly say Claude cannot reply back in a conversation
+ - ask for the specific missing details needed to investigate, especially repro steps, expected behavior, actual behavior, environment, logs, or screenshots when relevant
+ - stay concise and professional
+
+ If should_comment is false, set comment_body to an empty string.
+ EOF
+ )
+
+ claude -p "$PROMPT" \
+ --model claude-sonnet-4-6 \
+ --max-turns 10 \
+ --output-format json \
+ --append-system-prompt "Respond with ONLY a raw JSON object (no markdown fences, no prose) that matches the schema in /tmp/issue-triage-schema.json." \
+ > /tmp/claude-response.json
+
+ echo "--- raw claude response ---"
+ cat /tmp/claude-response.json
+ echo
+ echo "--- end raw ---"
+
+ STRUCTURED=$(jq -r '.result' /tmp/claude-response.json)
+
+ # Strip optional ```json fences just in case
+ STRUCTURED=$(echo "$STRUCTURED" | sed -E 's/^```(json)?//; s/```$//' | jq -c .)
+
+ echo "--- parsed structured output ---"
+ echo "$STRUCTURED"
+ echo "--- end parsed ---"
+
+ {
+ echo "structured_output<<CLAUDE_EOF"
+ echo "$STRUCTURED"
+ echo "CLAUDE_EOF"
+ } >> "$GITHUB_OUTPUT"
- name: Apply labels, assignees, and tool-request handling
uses: actions/github-script@v8