M6.4 module quiz
Ten questions. 80% to pass (8 of 10). Open book, unlimited retakes.
Answers are collapsed under each question. Answer first, then check.
Q1
A team-specific prompt beats a generic one because it encodes:
A. Better grammar and punctuation, so the model parses the request unambiguously
B. Scope, exclusions, time window, output format and context the team already knows
C. More detail throughout, since a longer prompt gives the model more
D. Model preferences, naming which model handles which kind of request
Show answer
Correct: B. Those five components are what a generic prompt makes the reader supply every time, and supplying them inconsistently is what makes results vary.
Q2
The test for whether a prompt is genuinely team-specific is:
A. Its length, measured in lines of instruction
B. Whether it names the team that owns and runs it
C. Whether another team could use it unchanged
D. Whether it produces a table as its output
Show answer
Correct: C. If they could, it is a generic prompt with a team’s name on it, and it belongs in the shared library rather than the team’s.
Q3
A reusable prompt is packaged as:
A. A saved chat thread in the assistant
B. A wiki page that the team copies from
C. A code comment in the shared repo
D. A skill or command file the client loads
Show answer
Correct: D. Which makes it versionable, reviewable and shareable through the same mechanisms as code, rather than living in someone’s history.
Q4
Parametrising a skill lets one definition:
A. Serve several teams or time windows without duplication
B. Run faster, by pre-resolving the tool calls it needs
C. Bypass RBAC, so the skill works for any caller at all
D. Use several different models within one invocation
Show answer
Correct: A. The alternative is five near-identical copies that drift apart, which is the maintenance failure the parametrisation exists to prevent.
Q5
The three sharing modes are:
A. Public, private and internal to the org
B. Read, write and admin, matching RBAC
C. Draft, in review, and published
D. Project-scoped, team-scoped, org-wide
Show answer
Correct: D. Choosing deliberately matters: an org-wide skill carries an implicit support obligation that a project-scoped one does not.
Q6
Quality gates for a shared skill should include:
A. Line count limits, so a skill stays short enough to review in one sitting
B. Model restrictions, naming which models the skill may be run against
C. A stated owner, a description of when to use it, and a review before publication
D. Approval from security, since a skill can reach any tool the caller can
Show answer
Correct: C. An unowned shared skill is the thing nobody updates when the underlying tool changes, and its silent failure is worse than its absence.
Q7
Skills should be versioned with:
A. Timestamps only, on each edit
B. Metadata plus a changelog
C. Git tags only, one per release
D. Sequential numbers, no metadata
Show answer
Correct: B. The changelog is what tells a user whether the output shape they depend on has changed, which is the question a version number alone cannot answer.
Q8
After a change to the underlying MCP tool surface, shared skills should be:
A. Spot-checked, because a skill referencing a changed tool fails silently
B. Rewritten from scratch against the newly changed tool surface
C. Left alone, since the client resolves tool names at run time anyway
D. Deprecated automatically, because any tool change invalidates them
Show answer
Correct: A. Silent failure is the characteristic mode: the agent produces something plausible from the tools that still work, and nobody notices the gap.
Q9
Deprecating a skill gracefully means:
A. Deleting it outright, so nobody can invoke the stale version by accident
B. Leaving it indefinitely, on the grounds that an unused skill costs nothing
C. Marking it deprecated with a pointer to the replacement, then removing it after a period
D. Restricting access to its original author, who is the only one still using it
Show answer
Correct: C. An abruptly deleted skill breaks whoever depended on it with no signal about what to use instead.
Q10
Output drift detection means:
A. Monitoring latency, so that a slow skill is caught before people stop using it
B. Periodically checking that a skill’s output still has the expected shape and content
C. Version-controlling outputs, so a change in results can be diffed over time
D. Comparing across models, to check the skill is not overfitted to a single one
Show answer
Correct: B. Skills degrade without failing, so the only way to notice is to look on a cadence rather than waiting for a report.
What’s next
Back to Team-specific prompts.