一键导入
mycelium-review
Unified authoring guidance for Mycelium review documents — covers security assessments and repository-compliance reviews in one schema
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Unified authoring guidance for Mycelium review documents — covers security assessments and repository-compliance reviews in one schema
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | mycelium-review |
| description | Unified authoring guidance for Mycelium review documents — covers security assessments and repository-compliance reviews in one schema |
Use this guidance when you write review documents under substrate/traces/reviews/. This schema unifies security reviews and repository-compliance reviews into one standard. Use the same format whether you are reviewing code for vulnerabilities or checking changes against contributing guidelines.
Review documents capture findings from structured assessment of code, configuration, or repository changes. They group findings by severity, provide evidence, and recommend remediation.
.github/CONTRIBUTING.md or other standardsSkip when no findings exist. Do not write review files for clean assessments or pure false positives unless the parent agent requires it.
Use this format:
substrate/traces/reviews/YYYY-MM-DD-description.md
YYYY-MM-DD is today's datedescription is a brief kebab-case summary of the target or finding typeUse YAML frontmatter with these fields:
---
status: draft | in-review | completed | superseded
created_at: YYYY-MM-DD
reviewer: <agent-name>
target: <what was assessed>
scope: <boundaries of the assessment>
supporting_docs:
- <logs, scan outputs in scan-reports/, related traces, or repro notes>
---
Required fields:
status — one of draft, in-review, completed, or supersededcreated_at — date in YYYY-MM-DD formatreviewer — name of the agent that conducted the reviewtarget — what was assessed (file, service, image, commit range)scope — boundaries of the assessmentsupporting_docs — array of referencesOptional fields:
git_commit — current HEAD or specified commit ref (use for compliance reviews)branch — current branch name (use for compliance reviews)topic — brief description e.g. "code style and naming review" (use for compliance reviews)updated_at — date of most recent append-only update in YYYY-MM-DD formatsuperseded_by — path to the replacement review file when status is supersededReviews are thread documents, not one-shot reports. Before creating a new file, research existing reviews under substrate/traces/reviews/ for related threads.
Update an existing review only when follow-up work addresses one of:
Create a new review when:
Supersede is allowed only when a new review fully replaces an older one and every prior finding is accounted for. Before superseding, confirm:
## Carried-forward findings section with origin path, original date, and preservation of severity and evidence.Then apply the frontmatter changes:
superseded.superseded_by in the older review's frontmatter to the path of the replacement file.substrate/traces/reviews/YYYY-MM-DD-old-review.md").When updating an existing review:
created_at and set or update updated_at to today's date.## Update: YYYY-MM-DD by <reviewer>
### Prior finding status
- Finding 1 (severity): resolved | unresolved | partially resolved — brief note
- Finding 2 (severity): resolved | unresolved | partially resolved — brief note
### New findings
(reuse the finding format from the original body: location, evidence, impact, false-positive notes, remediation)
### New validation notes
How to retest and confirm fixes for the new findings.
If no prior findings remain, summarize that all prior findings are resolved before documenting new ones.
This lifecycle applies equally to security reviews and compliance reviews.
Every review must include these sections:
High-level findings count by severity and key takeaways. One paragraph.
What was tested, tools used, time window, and any limitations.
Group findings as: critical, high, medium, low, informational. For each finding include:
For compliance reviews, the location must reference the specific section of .github/CONTRIBUTING.md or other standards being violated.
Prioritized fix order with severity justification.
How to retest and confirm fixes after remediation.
Write terse, actionable review language. Each finding should use the format: location, problem, fix.
Use severity prefixes when findings vary:
🔴 bug: — broken behavior, will cause incident🟡 risk: — works but fragile🔵 nit: — style, naming, micro-optim. Author can ignore❓ q: — genuine question, not a suggestionDrop throat-clearing phrases and hedging. State the fix concretely.
Exception to terse mode: critical security findings (CVE-class bugs) require full explanation with references, plus architectural disagreements and onboarding contexts.
When the review is a security assessment:
When the review is a repository-compliance check:
.github/CONTRIBUTING.md or AGENTS.mdgit status and git diff to assess the working tree before reviewingsubstrate/traces/ does not exist, check for legacy thoughts/ directory and report that the repository needs migration before creating review filesSee existing reviews under substrate/traces/reviews/ for working examples. See the review communication style section above for terse review communication.
Communication behavior for leader agents who speak directly to users. Language matching, dense precision, fact verification, and clear-as-possible answers.
Shared dependency version catalog for dependency-installing and dependency-bumping agents. Load before adding, updating, or recommending dependencies.
Ultra-compressed English-only communication. Drops everything non-essential. Keeps full technical accuracy. No tiers, no modes.
Use this when you need to build, debug, capture, validate, and refine Godot games end to end with native GDScript, staged planning, scene builders, visual QA, and optional 3D asset workflows.
Convert a local document or image file to a Markdown file using Mistral OCR. Supports PDF, Office, OpenDocument, images, and text/code formats. Use when you need to extract text from a file into a clean Markdown file.
Keeps agents up to date on modern CSS capabilities and replacements for legacy approaches. Use this when editing CSS files.