augchan42/inkstone / skills/deepen-task
deepen-task
VerifiedReframe a task or rewrite a prompt for greater rigor when asked for a deeper approach, expert methods, or more ambitious use of AI. Identify missing expert work, primary evidence, concrete methods, and independent validation. Use altitude and range only where they improve the user's goal. For reviewing an existing implementation plan, see plan-review-loop; for unclear marketing copy, see marketing-clarity.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.3 KB |
| references/examples-and-source.md | markdown | 6.0 KB |
references/examples-and-source.md
View on GitHub# Examples and Source Analysis
These are illustrative rewrites, not analyses that have been run. Adapt
methods to the actual evidence and constraints rather than copying them
into every task.
## Business analysis: investigate a churn claim
**Request:** "Make a quarterly review deck explaining why churn increased.
We think onboarding is the problem. Keep it to six slides."
**Diagnosis:** The deliverable is clear, but the explanation is assumed.
A stronger approach defines churn consistently, checks whether customer
mix changed, and tests the onboarding hypothesis against plausible
alternatives. Relevant event and billing records are more informative
than a chart of aggregate churn alone. Access to those records is an
input requirement, not something the prompt can grant.
**Rewritten prompt:**
> Produce a six-slide quarterly review explaining the increase in churn
> and the decision the evidence supports. Use the supplied billing and
> onboarding event exports; if only aggregate charts are available, state
> what they cannot establish and request the smallest relevant extracts.
> Define the churn numerator, denominator, and observation window before
> comparing acquisition cohorts and customer segments. Check whether the
> aggregate increase persists within comparable groups. Use SQL or a
> dataframe tool to make the calculations reproducible and compare the
> onboarding hypothesis with changes in customer mix and plan mix where
> the supplied fields permit. Treat associations as hypotheses about
> causes. Reconcile customer counts to the billing totals and check the
> explanation against another comparable period when the supplied history
> permits; otherwise state that limitation. Deliver six
> slides with source references and reproducible calculations in the
> slide notes, and the next decision or experiment. Stop at the supported
> explanation; do not invent a causal finding or build a monitoring system.
The notes make the calculations inspectable within the six-slide artifact.
An optional appendix should be proposed separately if the calculations
outgrow that format.
## Creative work: deepen a source-based painting prompt
**Request:** "Improve this ink painting prompt based on the supplied verse.
Keep the final prompt under 60 words."
**Diagnosis:** Better adjectives may polish an incorrect interpretation.
Start with the original verse and supplied translation, resolve the image
that carries its meaning, then choose composition and visual emphasis.
Philology contributes source accuracy; painting contributes composition.
Neither requires role-playing a panel of experts.
**Rewritten prompt:**
> Revise the supplied ink painting prompt into fewer than 60 words.
> Compare its depicted action and symbols with the original verse and
> supplied translation. Flag any ambiguity that changes the scene and
> avoid resolving it through invented historical context. Choose a clear
> focal action, framing, and depth that preserve the verse's emotional
> movement. Check the final scene against the source for contradictions
> and count its words. Deliver the prompt and a brief explanation of the
> consequential change. Label aesthetic choices as choices; visual
> coherence and source fidelity do not prove that an image will be liked.
This can lead into Inkstone's verse-to-prompt workflow when appropriate.
It does not replace that skill's established composition guidance.
## Boundary cases
- **"Make this two-line email friendlier; keep it two lines."** Edit the
email directly unless the user requests a deeper brief. No raw-data
investigation or audience study is needed.
- **"Review this implementation plan for missing steps."** Use
plan-review-loop for the review. Deepen-task addresses how to frame the
task, not the lifecycle of reviewing a written plan.
- **"Give me a deeper approach, then implement it in this repo."** Write
a brief diagnosis and proceed with authorized implementation.
- **"Give me a deeper analysis of these results."** Perform the analysis
using the available evidence and appropriate methods; do not substitute
instructions for a future analysis.
- **"Use only this PDF to rewrite my analysis prompt."** Work within that
evidence boundary. Explain what the PDF cannot establish; do not fetch
raw customer data or treat a desired finding as verified.
## Source and interpretation
- **Post analyzed:** [CG, 15 October 2026](https://x.com/cgtwts/status/2110571916973207906).
Its two attached screenshots highlight a depth-oriented prompt and the
four-lens framework. The short caption comments on the author's role;
the reusable method is in the screenshots and linked article.
- **Original announcement:** [Felix Rieseberg, 13 October 2026](https://x.com/felixrieseberg/status/2110023283169247623).
He describes himself as an engineering lead working on Claude. That
wording does not establish that he is Anthropic's sole lead engineer.
- **Primary source:** [How to be more ambitious: A tutorial for humans with models](https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/),
published 13 October 2026; read with both screenshots on 22 October 2026.
The useful distinction is between improving the task's formulation and
merely increasing its volume. The article connects that to four lenses:
speed, altitude, depth, and range. Its most directly reusable move is to
ask what professional work is missing, find the underlying evidence,
choose real methods, and define an outside check before rewriting a
request. It also warns that expert vocabulary and theatrical personas
can leave the method unchanged.
This is practitioner advice, not a controlled demonstration that a
particular prompt reliably improves every task. The skill therefore
requires task-specific evidence and checks rather than promising an
improvement. It treats aggressive targets as hypotheses to test and
cross-disciplinary work as concrete contributions, without importing
the article's expansive access assumptions or rhetorical encouragement
to keep pushing indefinitely.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 470k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |