removed duplicate

[sosahi]

Description

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• If adjusting docker-compose.yaml environment variables have you ensured those are mimicked in the Helm values.yaml file.

[greptile-apps]

Greptile Summary

This PR removes a duplicate Markdown metadata table from SKILL.md (whose fields are already captured in the YAML frontmatter) and strips all ./output.json write instructions from the query workflow, simplifying the skill so agents always answer in chat rather than writing a file.

• SKILL.md: Duplicate table removed, license key added to frontmatter, Query-turn row and Hard-limits bullet updated to say "Reply and then STOP" instead of writing ./output.json.
• references/query.md: General-use vs eval-harness callout removed; section renamed from "Write ./output.json directly from the hits" to "Compose your reply from the hits"; final STOP instruction updated accordingly.
• references/troubleshooting.md: Budget-violation recovery no longer tells the agent to write ./output.json.

Confidence Score: 4/5

Two no-narration instructions were not updated alongside the rest of the cleanup and still direct the agent to write a JSON file that no longer exists in the workflow.

The no-narration bullet in both SKILL.md and query.md still ends with "writing the JSON file," directly contradicting the updated Hard-limits and STOP instructions in the same files. An LLM agent reading the skill will receive conflicting guidance on whether to write a file or answer in chat.

skills/nemo-retriever/SKILL.md (line 34) and skills/nemo-retriever/references/query.md (line 14) — the no-narration bullets.

Important Files Changed

Filename	Overview
skills/nemo-retriever/SKILL.md	Removed duplicate Markdown table (info now in YAML frontmatter), added license field, and simplified Query-turn row and Hard-limits bullet to drop ./output.json — but the no-narration bullet still says "writing the JSON file".
skills/nemo-retriever/references/query.md	Removed general-use vs eval-harness callout and all ./output.json write steps; renamed section to "Compose your reply from the hits" — but the no-narration rule still ends with "writing the JSON file".
skills/nemo-retriever/references/troubleshooting.md	Removed stale ./output.json step from the budget-violation recovery instruction; no other issues.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Agent receives user prompt] --> B{Index exists?}
    B -- No --> C[Setup turn: read references/setup.md → build index → STOP]
    B -- Yes --> D[Query turn: read references/query.md]
    D --> E[Bash call 1: retriever query]
    E --> F{Fact in hits?}
    F -- Yes --> G[Compose reply from hits → STOP]
    F -- No / chart only --> H[Bash call 2: retriever pdf stage page-elements]
    H --> G
    E --> I{Error or empty?}
    I -- Yes --> J[Read references/troubleshooting.md → apply recovery]

Loading

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
skills/nemo-retriever/SKILL.md:34
Stale "writing the JSON file" instruction — the no-narration bullet still directs the agent straight to the JSON file write step, but the very next bullet and the query.md workflow no longer produce that file for general use. An agent following this skill will receive conflicting guidance: the Hard-limits section says "Reply and then STOP" while this line says "writing the JSON file."

```suggestion
- **No narration between tool calls.** Tokens you emit between calls become input + cached input for every later turn — quadratic cost. Go straight from reading the summary to composing your reply.
```

### Issue 2 of 2
skills/nemo-retriever/references/query.md:14
Same stale "writing the JSON file" text in the no-narration rule. This PR removes the `./output.json` write step for general use, but this sentence still instructs the agent to go straight to "writing the JSON file" — directly contradicting the renamed section ("Compose your reply from the hits") and the updated STOP instruction just below it.

```suggestion
**No narration between tool calls.** Do not write "Let me search…", "I'll now analyze…", "The retriever returned…", or any other commentary. Every assistant token you emit around the `retriever query` Bash call becomes input tokens (and cached input tokens) for every subsequent turn in this session — quadratic cost. Go straight from reading the summary to composing your reply. The only assistant text in a query turn should be the tool calls themselves.
```

_{Reviews (12): Last reviewed commit: "Merge branch 'main' into sohail/remove-d..." | Re-trigger Greptile}

[sosahi]

/nvskills-ci

[greptile-apps]

Broken reference — references/pitfalls.md does not exist

The table now directs error-recovery to references/pitfalls.md, but that file does not exist in the repository — only references/troubleshooting.md is present. Any LLM agent hitting an error and following this instruction will try to read a non-existent file and get no recovery guidance. Line 28 of the same file still points to references/troubleshooting.md, making the two references inconsistent.

Prompt To Fix With AI

This is a comment left during a code review.
Path: skills/nemo-retriever/SKILL.md
Line: 24

Comment:
**Broken reference — `references/pitfalls.md` does not exist**

The table now directs error-recovery to `references/pitfalls.md`, but that file does not exist in the repository — only `references/troubleshooting.md` is present. Any LLM agent hitting an error and following this instruction will try to read a non-existent file and get no recovery guidance. Line 28 of the same file still points to `references/troubleshooting.md`, making the two references inconsistent.

How can I resolve this? If you propose a fix, please make it concise.

[sosahi]

/nvskills-ci

[sosahi]

/nvskills-ci

[sosahi]

/nvskills-ci

[sosahi]

/nvskills-ci

[sosahi]

/nvskills-ci

[sosahi]

/nvskills-ci