customgpt-ai-rag:ask-agent

ask-agent

Ask a plain-language question about your project. Returns an AI-generated answer grounded in your files, with listed sources.

Rules

• ALWAYS read .customgpt-meta.json to get agent_id — never hardcode it
• Each call creates a fresh conversation session (stateless)
• ALWAYS present both the answer AND the sources — sources let the user verify accuracy
• If the agent has no processed documents yet, instruct the user to run /check-status first

---

Step 1 — Resolve API Key

Follow the lookup procedure in skills/_shared/api-key.md. Store the result as $API_KEY.

---

Step 2 — Read Meta File

Walk up from $PWD toward /, looking for .customgpt-meta.json. Extract agent_id, indexed_folder, and included_paths.

If not found:

"No agent found in this directory tree. Run /create-agent to set one up first."

---

Step 3 — Freshness Warning (Optional)

Check whether any files in the included paths were modified after the meta file's timestamp. For each path in included_paths (resolved relative to indexed_folder):

find "${indexed_folder}/${path}" -newer "${META_FILE_PATH}" -type f -not -path "*/.git/*" -not -path "*/node_modules/*"

If included_paths is missing or ["."], check indexed_folder itself.

If any files are newer, inform the user before proceeding:

"Note: {N} file(s) have changed since the last upload. Results may not reflect the latest changes. Run /update-agent to sync the changed files, or I'll continue with the current data."

Then continue regardless — do not stop.

---

Step 4 — Get the Question

If the user provided a question with the command, use it directly.

Otherwise ask:

"What would you like to search for?"

---

Step 5 — Create a Conversation Session

curl -s --request POST \
  --url "https://app.customgpt.ai/api/v1/projects/${AGENT_ID}/conversations" \
  --header "Authorization: Bearer ${API_KEY}" \
  --header "Content-Type: application/x-www-form-urlencoded" \
  --data "name=Claude+Code+Query"

Read the response and extract data.session_id as $SESSION_ID.

If the response indicates the agent has no content or is not ready, tell the user:

"The agent has no processed content yet. Run /check-status to see processing progress."

---

Step 6 — Send the Question

curl -s --request POST \
  --url "https://app.customgpt.ai/api/v1/projects/${AGENT_ID}/conversations/${SESSION_ID}/messages" \
  --header "Authorization: Bearer ${API_KEY}" \
  --header "Content-Type: application/x-www-form-urlencoded" \
  --data-urlencode "prompt=${QUESTION}" \
  --data "stream=false"

Read the response and extract:

• data.openai_response — the AI-generated answer

If openai_response is empty or null, the agent may still be processing. Tell the user:

"The agent returned an empty response. Your files may still be processing — run /check-status to verify, then try again."

---

Step 7 — Present Results

Prepare the output:

---

Answer

Use the {openai_response} to build the response, depending on what user prompted.

Sources

Use files listed in {openai_response} to build the table of sources in this format:

#	Source
1	{title or filename}
2	...

---

If sources are not present:

"No specific sources are provided. Agent either couldn't find the requested items, or it may still be processing your files. Run /check-status to confirm processing is complete."

After displaying results, consider whether any cited files are relevant to the current task. If so, use the Read tool to open them for full context before making code changes.