reflect — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited reflect (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Most "let me think about this" produces one perspective dressed up as several. This skill forces a real reflection by running multiple, structurally different methods against the same input, then surfacing where they disagree. The disagreement is usually where the real insight lives.
Core principle: a reflection that lands on the same answer the user came in with is not a reflection, it is a confirmation. Aim to find at least one thing that should change the user's mind, even if the final position holds.
/reflect <decision, idea, action, belief, or outcome>.Do NOT use for: brainstorming new options (use /innovate), prior-art search (use /research), or hardening an already-chosen direction (use /refine).
Restate the input as exactly one of these:
| Type | Example phrasing |
|---|---|
| Decision | "Should we ship X by Y?" |
| Belief | "I think X is true because Y" |
| Action taken | "We did X. Was that right?" |
| Outcome | "X happened. What does it mean?" |
| Idea or claim | "X is a good idea because Y" |
If the input does not fit one of these, ask once. The reflection method depends on the type.
Choose methods that are structurally different, not three flavors of the same lens. From this menu:
| Method | What it does | Best fit |
|---|---|---|
| First principles | Strip the question to facts that cannot be simplified further; rebuild the answer from those. | When the answer is leaning on inherited assumptions or analogies. |
| Steelmanning | Build the strongest possible version of each opposing position, including positions the user dislikes. Not strawmanning. | When there is real disagreement to resolve. |
| Pre-mortem | Assume the decision led to a bad outcome 12 months later. Write the autopsy. What went wrong? | Decisions, plans, commitments. |
| Pre-parade | The mirror of pre-mortem. Assume it went better than expected. What second-order effects did the user not plan for? | Decisions where the upside has not been thought through. |
| Inversion (Munger) | Instead of asking "how do I succeed?", ask "what would guarantee failure?" then avoid those things. | Spotting blind spots; risk reasoning. |
| Second-order thinking | "And then what?" Run the consequence chain at least two more steps than the user did. | Decisions whose first-order effects are obvious but whose downstream effects are not. |
| Stakeholder perspectives | Name 3 to 5 affected parties and reconstruct each one's view. Include parties without a voice in the room. | Decisions affecting other people; org or product choices. |
| Six Thinking Hats (de Bono) | Six perspectives in turn: white (facts only), red (gut and emotion), black (caution and risk), yellow (benefits), green (creative alternatives), blue (process and meta). | When emotion, creativity, and risk all matter and tend to get blurred. |
| Five Whys (Toyota) | Ask "why?" five times to drive past the surface explanation. | Outcomes and beliefs more than decisions. |
| Bayesian update | What evidence would update the user's belief? In which direction? By how much? | Beliefs and predictions. |
Defaults that work well for most prompts: first principles + steelmanning + pre-mortem + stakeholder perspectives. Add inversion when risk matters; add second-order thinking when downstream effects matter; add six hats when emotion is in play.
For each chosen method, write at minimum:
Do not blur the methods together. Each one is a separate paragraph or block. The point of using multiple methods is to see the same thing from incompatible angles.
Now compare the outputs. Specifically:
Pick exactly one of these and defend it:
A reflection that ends with "it depends" without specifying what it depends on is a failed reflection. Push to one of the three above.
# Reflection: [restated input]
## Type
[Decision / Belief / Action taken / Outcome / Idea]
## Methods used
[3 to 6 methods, named]
## Findings by method
### First principles
[What it says + non-obvious finding]
### Steelmanning
[What it says + non-obvious finding]
[... one block per method used ...]
## Tensions
- **Agreement:** [where methods converged]
- **Disagreement:** [where they conflicted, named explicitly]
- **What was missing:** [what the original framing did not address]
## Position
**[HOLDS / CHANGES / DEFERRED]**
[2 to 4 sentences defending the call. If DEFERRED, name the specific question and the cheapest way to answer it.]
## What I would do next
[1 to 3 concrete actions]| Failure | Fix |
|---|---|
| Picked three methods that all favor the same lens (e.g. three risk-focused) | Mix methods that look at the question from incompatible angles. |
| Steelmanned weakly (built a strawman in disguise) | A real steelman makes the opposing view sound persuasive even to people who hold the user's view. If the steelman is easy to dismiss, redo it. |
| Skipped the tension step | Tensions are the value. Do not collapse them into a polite synthesis too early. |
| Ended on "it depends" with no specifics | "Deferred" requires naming the question and the resolution path. |
| The reflection confirmed the user's prior with no challenge | Aim to find at least one thing that should update the user's view, even if the final position holds. If you cannot, say so explicitly. |
| Used six hats by default for everything | Six hats is for when emotion and creativity matter. For most decisions, first principles + steelmanning + pre-mortem is faster and sharper. |
| Pre-mortem was generic ("things might go wrong") | A real pre-mortem is concrete. Name the specific failure mode, the leading indicator, and how to detect it early. |
/design-ai-experiment (in the public bundle)./research (in the private bundle)./innovate (in the private bundle).~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.