Ten-Principle Thinking Loop
A structured ten-stage loop that helps Claude analyze complex problems, challenge assumptions, and produce actionable conclusions.
The skill explicitly says it does not modify files or execute mutations, and declares Read, Grep, Glob, and Bash; however, Bash remains broad and there is no per-step confirmation, least-privilege rationale, data-flow disclosure, sensitive-data guidance, or rollback procedure. Repository SECURITY.md is useful repository-level context but does not close the skill-specific gaps, so points are deducted.
The ten stages, invocation conditions, and anti-patterns are internally consistent, and failure feedback is included conceptually. However, there is no executable implementation, abnormal-input handling, or key-path reproduction evidence. Under static calibration the score cannot exceed 10, so it is rated 8.
The audience, scenarios, trigger phrases, and non-fit cases for trivial tasks are clearly stated. Input/output boundaries remain abstract, there is no explicit Chinese-language or mainland-China environment guidance, and execution depends on the model carrying out the stages, so points are deducted.
Frontmatter, version v1.9.0, stage structure, usage instructions, anti-patterns, composition notes, and references are well organized. Repository context also supplies an MIT license, CI, SECURITY.md, and a maintenance contact. The selected skill lacks a changelog, explicit ownership/update path, FAQ, and representative complete output example; some referenced files are not included in the supplied material, so points are deducted.
The skill provides directly reusable prompts for a ten-stage reasoning loop and is plausibly useful for architectural decisions, audits, and post-mortems. It lacks verified representative outputs, a defined output format, and evidence of marginal benefit over manual reasoning; quality depends on model execution. Static calibration caps this at 7, so it is rated 6.
The skill's rules and triggers are auditable, while the repository CI and tests provide general maintenance signals. The supplied tests do not cover the think skill's key paths, and there is no third-party execution evidence or corroborating validation, so static verifiability is rated 4.
- The skill claims not to execute mutations but permits Bash; adopters should verify whether the actual host can run commands or cause external side effects.
- Output format is unspecified, so results for complex tasks may vary with model execution.
- Documentation is primarily English and provides no Chinese terminology, examples, or mainland-China network guidance.
- Referenced audit documentation and other skill appendices were not supplied, so their contents cannot be statically verified.
What does this skill do, and when should you use it?
Think is a meta-skill in claude-obsidian for non-trivial problems such as architectural decisions, post-mortems, audits, ambiguous requests, and stakeholder trade-offs. It moves through ten stages covering observation, listening, critical analysis, lateral and systems thinking, empathy, humility, creation, and growth. The framework emphasizes reading evidence first, examining bias, applying first-principles reasoning, and learning from the resulting artifact. It does not modify files or execute mutations.
The skill can read project files and raw inputs, then guides Claude through external observation, internal metacognition, active listening, critical processing, lateral connection, system orchestration, emotional and intuitive consideration, intellectual humility, generative output, and iterative growth. Its prompts address reading before changes, recording bias, interpreting failure messages, checking integration, identifying omissions, and ending with a feedback loop. The resulting artifact may be a recommendation, decision, audit report, or other problem-solving output.
- An architect comparing designs needs to evaluate constraints, alternatives, and integration risks before committing to an approach.
- An engineering team conducting a post-mortem needs to examine evidence, bias, failure causes, and improvements for the next cycle.
- An auditor reviewing a system or skill needs a consistent methodology spine for finding omissions and cross-cutting issues.
- A user with an ambiguous or multi-stakeholder request needs to clarify intent and absorb signals before deciding.
- A team preparing a high-stakes release needs to test whether its evidence supports the verdict and avoid inflated claims.
What are this skill's strengths and limitations?
- Provides a clear ten-stage method spanning evidence collection, bias awareness, analysis, creation, and feedback.
- Fits architectural decisions, audits, post-mortems, ambiguous requests, and multi-stakeholder trade-offs.
- Its ACCEPT stage explicitly requires honest findings and verdicts, helping resist sycophancy.
- It does not modify files or execute mutations, making it suitable as an analysis discipline.
- It is a reasoning framework rather than an executor; results still depend on the model and the quality of the inputs.
- It is intentionally unsuitable for trivial tasks, where the full loop can add unnecessary ceremony.
- The SKILL.md does not document a standalone installation path; installation is documented for the repository or plugin collection.
- No standalone test suite or cross-platform verification evidence is provided for this skill.
How do you install this skill?
The repository provides think as one of 15 bundled skills and does not document a separate think-only installation. To install the public collection, run:
git clone https://github.com/AgriciDaniel/claude-obsidiancd claude-obsidianbash bin/setup-vault.sh
Alternatively, in Claude Code run claude plugin marketplace add AgriciDaniel/claude-obsidian, followed by claude plugin install claude-obsidian@agricidaniel-claude-obsidian.
How do you use this skill?
Use /think <problem statement> before a non-trivial problem, or use a listed trigger such as think this through, 10-principle review, audit my thinking, or systematic thinking. The skill recommends moving through the ten stages in order on the first pass and looping back when new information emerges. It is not recommended for single-line fixes or trivial lookups.