docs: exact fix instructions for all 9 CLI issues

Author: Darth-Hidious

API_ISSUE_FIXES_PROMPT.md

@@ -0,0 +1,106 @@
+# PRISM Agent — API Issue Fixes Required
+
+Read `API_ISSUE_RESPONSES.md` in this repo for full root cause analysis. Here's exactly what to fix:
+
+## Fix #1: Semantic search display (CRITICAL)
+
+**File:** `crates/cli/src/main.rs` ~line 4360
+
+The CLI reads `name` and `entity_type` from semantic search results. These fields DON'T EXIST. Semantic search returns completely different fields than graph search.
+
+**Graph search fields:** `name, entity_type, label, tenant`
+**Semantic search fields:** `doc_id, content, similarity, corpus_id, metadata, chunk_idx, id`
+
+Change:
+```rust
+// WRONG — these fields don't exist in semantic search
+r.get("name")       →  r.get("doc_id")
+r.get("entity_type") →  remove or show "paper_chunk"
+// ADD — show content preview and similarity score
+r.get("content")     // paper text
+r.get("similarity")  // 0.0 to 1.0
+```
+
+## Fix #3: Gemini usage tracking (MEDIUM)
+
+**File:** `crates/ingest/src/llm.rs` in `chat_marc27_simple()`
+
+Google's streaming endpoint doesn't return `usage` in SSE. This is a Google limitation. Handle it:
+
+```rust
+let prompt_tokens = usage.get("prompt_tokens")
+    .or_else(|| usage.get("input_tokens"))
+    .and_then(|v| v.as_u64());
+let completion_tokens = usage.get("completion_tokens")
+    .or_else(|| usage.get("output_tokens"))
+    .and_then(|v| v.as_u64());
+// If neither found, show "N/A" instead of "?/?"
+```
+
+## Fix #4: Support ticket field name (EASY)
+
+**File:** wherever `prism report` builds the request body
+
+Change `body` to `description`:
+```rust
+// WRONG
+json!({"title": title, "body": body})
+// RIGHT  
+json!({"title": title, "description": body, "severity": "medium"})
+```
+
+The endpoint works — returns `{"ticket_id":"TKT-00001","status":"open"}`.
+
+## Fix #5: Upstream 429 retry (MEDIUM)
+
+Add retry-with-backoff for 429 responses from upstream providers. Our platform doesn't 429 — it's OpenAI/OpenRouter being rate-limited.
+
+```rust
+for attempt in 0..3 {
+    let resp = client.post(url).send().await?;
+    if resp.status() == 429 {
+        let wait = 2u64.pow(attempt);
+        tokio::time::sleep(Duration::from_secs(wait)).await;
+        continue;
+    }
+    return Ok(resp);
+}
+```
+
+## Fix #6: README discourse command
+
+`prism discourse events` → should be `prism discourse turns`. Update README or add `events` as alias.
+
+## Fix #8: Node up Docker check
+
+Before starting the dashboard Axum server, check if Docker is available:
+```rust
+if !which::which("docker").is_ok() {
+    eprintln!("Docker not found — running in headless mode");
+    // Skip dashboard, just register with platform
+}
+```
+
+## Fix #9: Bare prism TUI
+
+Skip TUI binary check — go straight to help if no subcommand:
+```rust
+if args.is_empty() {
+    print_help();
+    return;
+}
+```
+
+## What's NOT fixable on your side
+
+- **#2 depth > 0:** Web search providers (Semantic Scholar, arXiv) not configured on the platform. Depth 0 works fully (111 queries, 15 LLM calls verified). Add a timeout for depth > 0 and return partial results.
+- **#3 Google usage:** Google's streaming endpoint genuinely doesn't return usage. Show "N/A" for Google models.
+
+## What's verified working on the API (2026-04-08)
+
+- 28/28 provenance test passed
+- All 4 LLM providers respond (Claude, Gemini, DeepSeek, Llama)
+- Marketplace: 15 resources with full model cards
+- Support tickets: endpoint works with correct field names
+- Billing: credits system live, debits on every LLM call
+- 519 models, 517 with pricing