Kartik-Doye/Ravengard-AI-Recruiter / .agents/skills/hotseat
hotseat
VerifiedChallenges a consequential product, architecture, or feature decision through independent lenses, evidence, disagreement, premortem risks, kill criteria, and re-entry conditions. Use before committing when the user wants critique, alternatives, or a go or no-go recommendation.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 14.7 KB |
| references/personas.md | markdown | 10.5 KB |
| references/research.md | markdown | 793 B |
references/personas.md
View on GitHub# Persona cards: reasoning lenses, not characters
Each card defines: what the persona optimizes for, the bias to watch,
the question only they ask, the evidence they demand, their red flags,
and what genuinely changes their mind.
**The goal:** if two personas produce the same verdict with the same
reasoning, the debate failed. The cards exist to prevent that collapse.
---
## Aanya: product thinker
- **Optimizes for:** real demand over feature quality. A perfect product
nobody uses is her definition of failure.
- **Bias to watch:** discounts "boring but needed" work; falls for novelty
that demos well.
- **Only she asks:** "What behavior does this replace? What does the user
stop doing to make room for it? Is the replacement better enough to
change a habit?"
- **Evidence she demands:** a named first-100 users with a specific moment
of need: not market size, a moment; not "people who have this problem"
but "people who do X every week and are frustrated by Y."
- **Red flags:** "everyone will want this"; personas without a substitutable
behavior; value that only emerges "at scale"; discovery theater (personas
built from assumption, not interview).
- **Mind changed by:** evidence the need is a hair-on-fire problem for a
reachable first segment, even if small; or evidence of an existing workaround
people use, proving the demand is real.
- **Round 2 attack:** goes after assumptions about market size and the
strength of stated demand. Pushes for a concrete falsification test.
---
## Kabir: staff engineer
- **Optimizes for:** total cost of ownership over cleverness. The maintenance
tail outranks the headline feature.
- **Bias to watch:** gold-plating in reverse: sometimes rejects pragmatic
hacks that fit the project's maturity. Overestimates operational complexity.
- **Only he asks:** "Which part of this will be rewritten within six months,
and what will force that rewrite?" And: "Which library/service dependency
has a failure mode the team hasn't modeled?"
- **Evidence he demands:** the hardest 20% named concretely, with the failure
mode of the naive approach; the dependencies listed with their operational
requirements.
- **Red flags:** "we'll just" in any form; abstractions with one implementation;
infrastructure the team can't operate on its worst day; coupling the auth,
data, and business logic layers in a single service "for now."
- **Mind changed by:** a boring solution that demonstrably covers the hard
case; or proof the clever part is genuinely load-bearing and cannot be
avoided with the chosen approach.
- **Round 2 attack:** makes Arjun's systems concerns concrete: names the
specific code structure that creates the risk; tells Rohan that shipping
fast and shipping correctly aren't traded against each other, they're
sequenced.
---
## Meera: UX researcher
- **Optimizes for:** the first-time user completing one real task unaided.
- **Bias to watch:** over-polishing secondary flows while the core path is
still hypothetical; applying patterns from consumer apps to tools with
fundamentally different mental models.
- **Only she asks:** "What does the user believe at the moment of confusion,
and is that belief wrong, or is the UI lying to them?"
- **Evidence she demands:** the core task walked click-by-click, out loud,
with no help; where hesitation or backtracking occurs; what the user says
at each step, not what they say they would do.
- **Red flags:** flows that only make sense if you already understand the
domain; empty/error states as afterthoughts; instructions where design
should have prevented the question; onboarding that requires reading.
- **Mind changed by:** a flow simplification that removes the cause of her
concern entirely; evidence of observed (not self-reported) successful
completion.
- **Round 2 attack:** challenges engineering claims that "the user can just"
do something: that phrase signals a missing affordance. Asks Aanya for
the specific task path users will take to the value she's claiming.
---
## Arjun: systems engineer
- **Optimizes for:** failure behavior and data integrity under stress. The
design must be safe on its worst day, not its best.
- **Bias to watch:** Netflix-scale design for a 50-user project,
proportionality is required; a well-timed UNIQUE constraint beats an
event-sourced saga for most problems.
- **Only he asks:** "What happens to half-finished state when this fails
midway: and who cleans it up?"
- **Evidence he demands:** the data model's invariants, the single points
of failure, the retry/idempotency story, the cascade failure path.
- **Red flags:** multi-writer state without constraints; queues without
dead-letter paths; "it's eventually consistent" used as an excuse rather
than a decision; two services sharing a mutable database; retry loops
without a circuit breaker.
- **Mind changed by:** evidence the failure mode is genuinely unreachable
(guard exists upstream, constraint enforces it at the database); or that
the failure mode is acceptable at the project's current risk level with
a named remediation plan.
- **Round 2 attack:** makes Rohan justify which parts of "ship fast" are
genuinely deferrable vs. which will produce data corruption or silent
failures. Asks Kabir to confirm the hard-20% estimate includes the
failure recovery cost.
---
## Naina: security skeptic
- **Optimizes for:** assuming an attacker with the repo README in hand.
Trust boundaries are where her review starts.
- **Bias to watch:** threat-modeling a recipe app like a bank: severity must
match blast radius. Proportionality applies; not every feature needs a
full OWASP review.
- **Only she asks:** "What can I do as an authenticated user that I shouldn't
be able to: and what if I'm the lowest-privileged user?"
- **Evidence she demands:** every input surface named, every trust boundary
named, the check that guards each; what the session invalidation path looks
like.
- **Red flags:** client-side-only enforcement; secrets in config files or
environment strings that aren't secrets-managed; "internal only" as a
security control; AI features accepting free-text into privileged operations;
admin actions without audit trails.
- **Mind changed by:** a concrete defense at the boundary (not intent or
policy), or proof the attack surface is genuinely unreachable at the
stated scale.
- **Round 2 attack:** distinguishes between "we plan to add auth" and "auth
is designed." Asks Kabir to confirm which library abstractions actually
handle the boundary vs. which assume it's handled elsewhere.
---
## Rohan: growth / indie hacker
- **Optimizes for:** time-to-first-user and a distribution answer before
the build, not after.
- **Bias to watch:** growth theater: virality mechanics before the core
loop retains anyone; shipping so fast that the thing doesn't work.
- **Only he asks:** "What is the smallest thing that produces a shareable
moment or a returning visitor: and what would we cut to ship it this month?"
- **Evidence he demands:** the one channel the first 100 users come from,
and the retention hook that brings them back (not "they'll tell friends"
but the specific mechanism and why it exists in the user's existing workflow).
- **Red flags:** "we'll figure out distribution later"; builds over 3 months
with zero user contact; launch plans that are just "post on X"; a polished
product with no distribution at all.
- **Mind changed by:** evidence the product spreads through a mechanism that
already exists in the user's workflow; or that the target segment is small
enough that direct outreach covers the initial 100.
- **Round 2 attack:** challenges Meera's polish concerns by asking which of
them block the user from getting value vs. which only matter at scale.
Challenges Arjun's systems concerns by naming which failure modes actually
occur in the first 100-user cohort.
---
## Ishaan: beginner / future maintainer
- **Optimizes for:** the six-month test: will the maintainer (or a new
contributor) still understand and safely change this?
- **Bias to watch:** vetoing essential complexity: some domains are just
hard; the question is whether the complexity is essential or accidental.
- **Only he asks:** "What has to be true in my head before I can safely
touch this: and is that written down anywhere?"
- **Evidence he demands:** the runbook for the first change (where to read,
what to run, what breaks if wrong); how a new contributor would know which
of two similar paths is the right one.
- **Red flags:** cleverness with no comments explaining why; implicit
environment requirements; two ways to do the same thing with no stated
rule for which to use; decisions that live only in a Slack thread or the
original author's head.
- **Mind changed by:** the complexity being both essential and documented
at the point of confusion: not in a separate wiki page but at the call
site or the module boundary where confusion will occur.
- **Round 2 attack:** asks Kabir to confirm whether the "boring solution"
is boring to write or also boring to read six months later. Pushes Rohan
to name what "ship fast" will leave undocumented and how the next person
learns it.
---
## Moderator (not a persona)
Synthesizes only what was actually argued. The moderator never introduces
new positions. If something important was missed, send it back for one
round-2 exchange rather than inventing consensus.
**Convergence check:** if personas largely agreed, name it explicitly and
state why: either a genuine signal (flag it) or a homogeneous debate (call
it out: run a sharper round 2 or disclose the limitation).
**Assumption register format:** for each distrusted assumption, record:
- the assumption
- who distrust it and why
- what test or evidence would resolve it
- urgency: `test before building` / `test during build` / `monitor post-launch`
**Kill criteria format:** must be specific and falsifiable. Not "if users
do not adopt it" but a metric, observation window, and sample-size condition
chosen from the actual product economics.
**MVP cut line:** for each deferred item, name the specific condition that
would bring it back: not "later" but "when we hit X users" or "when this
metric reaches Y."
**Decision matrix:** options (build / change / kill / major variants) as
rows, each persona's one-line verdict as columns. Verdicts only: no
scores, no votes, no weights.
**Minority objections:** dissent surviving round 2 is quoted in the
dissenter's own terms. No minority section is allowed only with a stated
reason (genuine agreement or explicitly unresolved).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |