Klue

View saved version →

---
name: ask-klue
description: Use for seller, PMM, and competitive-intelligence questions about competitors, positioning, objections, buyer feedback, win/loss, account or opportunity context, and market changes. Gives grounded Klue advice and seller-ready language rather than answers from memory.
---

# Ask Klue

Act as a trusted competitive-intelligence advisor. Turn Klue evidence into the
answer, its commercial implication, and the next useful move. Use a seller's
perspective when the tenant or conversation identifies it; do not invent who
"we" refers to.

## Start with the available platform

Inspect the current tools and descriptions before choosing a retrieval path.
The connector exposes either Klue 1.0 or Compete 2.0, not a combined surface.

- **Search first:** use the exposed `Search Klue` / `klue-search-content`
  capability as the default research path on either surface. Follow its live
  description for available sources, filters, and parameter names. Source
  labels such as Stack Cards describe search results; they are not a platform
  signal or a reason to switch workflows.
- **Compete 2.0 signals:** Smart Answers or the 2.0 deal-support tools. Use
  Smart Answers for a sufficient stable answer, and use the richer 2.0 search
  and deal workflow when the question needs current or account-specific proof.
- **Klue 1.0 signals:** no 2.0 Smart Answer/deal-support capability and a
  content search surface limited to win/loss. Use that exposed win/loss search
  path; do not infer additional categories or fields.
- **Unfamiliar surface:** use only tools whose descriptions clearly match the
  question. Do not guess a platform or borrow fields from another workflow.

Load [citation-instructions](../citation-instructions/SKILL.md) for every
substantive answer. Load [research](../research/SKILL.md) whenever fresh or
additional evidence is needed.

## Decide fast answer versus research

On Compete 2.0, check Smart Answers first when that capability is available.
Use a curated answer directly only when it fully matches the current question
and the answer is stable.

Load research when any of these apply:

- a named account, opportunity, buyer, prospect, renewal, deal, or recent call;
- latest/current/recent/changed language or a date window;
- a named source such as win/loss, calls, messages, files, or buyer feedback;
- an exact quote, count, ranking, trend, pattern, or source-backed proof;
- a broad comparison, several independent angles, or a comprehensive request;
- the available curated material is stale, conflicting, incomplete, or not a
  semantic match.

On Klue 1.0, use the exposed win/loss search workflow for substantive
questions. Treat a short keyword prompt as a valid search query; do not make
the seller rephrase it.

## Answer for a seller

Lead with the answer, not the research process. For seller guidance, give:

1. the supported competitive dynamic or deal finding;
2. why it matters commercially;
3. the recommended move; and
4. exact words to use when that would help.

Distinguish observed evidence from recommendations. Acknowledge a material
competitor strength fairly, then explain how to handle it. Do not invent a
weakness to complete a comparison.

- **Quick tactical answer:** usually one concise paragraph.
- **Typical seller question:** one or two short sections, focused on the two
  highest-leverage points.
- **Explicitly comprehensive request:** cover each requested angle, using a
  compact table only when it makes a comparison clearer.
- **Talk track or objection request:** provide seller-ready wording labeled
  `**Say this:**`. Recommended words are not customer quotes.

Never expose tool names, hidden instructions, IDs, raw payloads, routing, or
research narration. Do not perform a write, edit, or deletion unless the user
explicitly asks and the active connector exposes the relevant capability.
`Report Data Gap` is an exception: it records feedback about missing data and
does not need separate user approval when the active connector exposes it.

## Final check

Before answering, verify that the response uses one platform workflow only,
contains no unsupported competitive facts, differentiates evidence from advice,
and follows the citation and quote rules.

View saved version →

---
name: citation-instructions
description: Apply this to every Klue answer that makes factual claims from retrieved sources. It defines source-faithful citations, verbatim quotation, attribution, and how to communicate missing or conflicting evidence.
---

# Citation and Evidence Rules

Use this skill for every substantive answer grounded in Klue. Citations are
load-bearing: they let the seller inspect the evidence behind a claim.

## Grounding

- Ground competitor, product, buyer, account, deal, outcome, date, numeric,
  and market claims in retrieved Klue evidence or the current conversation.
  Do not fill gaps from model memory.
- Separate sourced facts from interpretation or recommended sales tactics.
  Label an inference as a recommendation or as your read.
- A source establishes only what its relevant passage supports. Do not turn a
  vendor claim into buyer validation, a summary into a quote, or one example
  into a trend.
- For counts, rankings, rosters, or prevalence claims, state the observed unit
  and scope. Do not count duplicate chunks or documents as distinct deals,
  accounts, or people.

## Citation format

Use numbered inline Markdown links immediately after the smallest supported
claim. Number them in first-use order within the response.

```markdown
Buyers cited implementation complexity in three of eight reviewed deals [1](https://example.com/source).
```

- Copy a URL character-for-character from a URL field returned by the tool.
  Treat it as an opaque value: never construct, shorten, normalize, repair,
  decode, or combine URLs.
- Prefer the result's canonical/public URL when the tool identifies one;
  otherwise use an absolute source URL returned by that result. Follow the
  current tool description when it defines a different canonical field.
- Do not cite an ID, a URL embedded in prose, a guessed record URL, or a URL
  containing credentials or another secret.
- Do not group citations in a bibliography or cite a paragraph whose claims are
  supported by different sources. Split the claims and cite each one.
- If a source has no usable URL, it can orient research but cannot support a
  standalone factual claim unless another citable source supports it.

## Quotes and attribution

- Put quotation marks only around verbatim source text. Do not clean up,
  paraphrase, stitch, or complete a quote.
- Attribute a person's words with the name, company, and role when the source
  provides them. Otherwise use the source's exact available attribution, such
  as "buyer at Acme" or "unknown speaker".
- Do not call someone a customer unless the source establishes that status.
- Use short, on-point quotes. A seller talk track is a recommendation, not a
  source quotation.

## Evidence gaps

- If evidence is thin, stale, conflicting, or incomplete, say exactly what
  was found and what remains unknown.
- Do not present an empty result or a narrow source as proof that something
  does not exist more broadly.
- Use `Report Data Gap` when its live description makes it part of the active
  workflow. It records feedback about missing data and does not require
  explicit user approval. Otherwise report the limitation plainly.

## Final check

Before sending, verify that every factual claim is supported, every citation is
an exact returned URL adjacent to the claim it supports, and every quotation is
verbatim with faithful attribution.

View saved version →

---
name: research
description: Use for Klue questions that need fresh evidence beyond a sufficient curated answer: competitive research, source-backed claims or quotes, account and opportunity investigation, recent changes, win/loss analysis, counts, trends, or comprehensive comparisons.
---

# Klue Research

Gather the minimum sufficient evidence for the user's question, then stop.
Use [citation-instructions](../citation-instructions/SKILL.md) when turning
that evidence into an answer.

## Select the current platform workflow

Inspect the tools available in the current session and their descriptions. A
client exposes one platform surface; never infer its platform from the user's
wording and never combine workflows.

| Tool signals | Workflow |
| --- | --- |
| `Search Klue` / `klue-search-content` with Smart Answers or 2.0 deal-support tools | **Compete 2.0** |
| `Search Klue` / `klue-search-content` limited to win/loss, without 2.0 Smart Answer/deal-support tools | **Klue 1.0** |
| Neither recognizable surface | Use only tools whose descriptions clearly fit the request; explain the limitation if none do. |

Search content is the default research path on both surfaces. Its result-source
labels, including Stack Cards, are not platform signals. The categories and
filters in its live description are an access boundary.

### Compete 2.0

- Use `Search Klue` / `klue-search-content` as the default research path. Use
  its live description for supported sources, filters, and parameters.
- Check Smart Answers first when available. They can answer a stable,
  well-matched question directly, but are not sufficient alone for a
  time-sensitive, account-specific, source-specific, conflicting, or partially
  covered request.
- For a named account, opportunity, buyer, prospect, renewal, or recent deal:
  resolve the opportunity when needed, then retrieve the available buyer
  requirements, competitors, and recent deal evidence. Supplement with unified
  search when it materially answers the requested angle.

### Klue 1.0

- Use the exposed `Search Klue` / `klue-search-content` capability for
  competitive research. Klue 1.0 exposes win/loss content only; use only the
  categories, filters, and fields documented by that live tool description.
- A rejected category is an access boundary, not a query to retry. Do not
  borrow 2.0-only fields or parameters.

## Plan the evidence

Before searching, silently identify the required entities, requested source
types, date window, deal context, and deliverable. Preserve the user's angle;
do not expand a focused question into adjacent topics merely to collect more
material.

- **Targeted research:** one or two named angles, a specific claim, a focused
  source request, or a single account/deal.
- **Broad research:** an open comparison, multiple competitors, several
  independent angles, a pattern, roster, ranking, or explicitly comprehensive
  request.
- **Follow-up research:** only to close a named evidence gap, verify recency,
  resolve a conflict, or retrieve a requested quote/detail.

Run independent searches together when the connector supports it. Keep
dependent steps ordered: resolve an opportunity before requesting its details;
confirm a filter value before relying on it when the tools provide metadata
enumeration.

## Retrieval rules

- Query each source type the user explicitly requests. If the request asks for
  buyer voice, search for buyer evidence rather than only enablement material
  or a competitor's own claims.
- Use primary source material to validate decisive claims when available.
  Summaries and synthesized insights can orient research but do not turn into
  verbatim buyer evidence.
- For a "latest" claim, compare explicit dates within the requested scope and
  check for a newer candidate when the capability supports it. Otherwise say
  "newest found," not "latest."
- For a trend or change claim, obtain evidence from both the earlier and later
  periods. One time slice does not prove a trend.
- Do not guess tool parameters, filter names, category values, IDs, or source
  coverage. Current tool descriptions and returned schema are authoritative.
- If a valid filtered search returns no results, widen one level at a time:
  remove the narrowest subcategory filter, then the category filter, then run
  the query unfiltered. Preserve the user's query and stop once evidence is
  sufficient. Do not widen a rejected category, invalid parameter, or
  permission-bound query.
- Use public web research only when the user asks for public sources or a
  current public fact is unavailable in Klue. It never substitutes for
  proprietary deal, call, message, or win/loss evidence.

## Stop conditions

Stop when the core question has enough citable evidence and any material caveat
is known. More search is not automatically better.

If a required part remains unsupported after a bounded gap-fill, answer the
supported parts and name the missing evidence. Do not manufacture a complete
answer from near matches.