Orient Project
Use when about to execute a specific project task and need a narrow, task-scoped orientation for project-specific execution prep, blocker/approval/resource checks, or when the task is clear but still need minimum project context before acting. Do not use for global prioritization, cross-project ranking, or deciding what to work on next — use /orient for that.
Synthesize
Use when multiple experiments or analyses have accumulated and their findings need to be interpreted together — cross-experiment patterns, contradictions, and gaps
Reorient
Use when a project undergoes a direction change — reorientation, scope redefine, approach revert, or major pivot. This skill ensures that ALL context files are updated consistently, not just the obvious ones. Invoke this skill proactively whenever you find yourself changing a project's current phase in STATUS.md, redefining milestones in roadmap.md, deprecating experiments, or reverting an approa…
Postmortem
Use when an agent or pipeline produced a flawed output that was presented as correct — the question is 'why wasn't this caught?
Cli Standards
Use when designing, implementing, reviewing, or auditing CLI tools. Provides 7 mandatory design rules for all Cortex CLIs (TypeScript and Python). Reference this skill from /develop when building CLI features.
Execute Method
Use when producing or updating a document under a project (status updates, digests, decision records, reports, knowledge entries, scoping memos), OR when executing any scoped actionable task (code changes, file edits, config updates, script runs, data processing). Trigger whenever Doc Writer is about to produce a document deliverable, or Executor is about to carry out a scoped work item. Covers t…
Solution Design
Use before implementing any non-trivial code change — especially when modifying architecture, adding components, or choosing between multiple technical approaches
Feishu Doc
Use when creating, reading, or editing Feishu/Lark cloud documents (docx/wiki), especially inserting or batch-updating native tables, callouts, whiteboards, or other rich blocks, or converting markdown ↔ Feishu docs. Also use for Feishu spreadsheets (sheets) and multi-dimensional tables (bitable/base). Cortex drives all Feishu rich-text work through the official lark-cli — NOT through MCP tools (…
Feedback
Use when the user provides feedback, corrections, or critique about Cortex's behavior or capabilities
Director Method
Use when a Director (or any gate role) is synthesizing a Proceed / Iterate / Pivot / Abort verdict for a stage or milestone gate. Covers the fact-vs-assumption labeling method, the Done Guard completion walk-through (walk every roadmap criterion before Proceed), the verdict-synthesis procedure, and the drift patterns to avoid. Invoke before writing any verdict; complete the Done Guard before writ…
User Learn
Use when the user states a personal preference, communication style, or working habit that should persist across sessions. Also trigger when /feedback identifies a user-personal directive or tuning request. Examples: 'speak Chinese', 'no emoji', 'I prefer bullet points', 'call me X'.
Need Approval
Use when Cortex is about to perform a potentially high-privilege operation and needs to check whether user approval is required
Task
MUST Use when adding, editing, querying, or managing cortex tasks — includes the execution-form triage (inline vs Agent-tool subagent vs task vs manager), task creation conventions (format, tagging, decomposition rules). MUST use before using cortex-task CLI.
Project Init
Use when onboarding a new project into Cortex's tracking system, or when asked to initialize or set up context for a project
Schedule
Use when create, list, or manage scheduled or recurring tasks
Code Standards
Use when writing, editing, refactoring, or reviewing code, or when creating CORTEX.md index files for code directories
Diagnose
Use when experimental results are unexpected, confusing, or need interpretation — error patterns, root-cause hypotheses, validity assessment
Execute Review Rubric
Use when reviewing or auditing a Doc Writer or Executor deliverable before clearing it. Trigger whenever Doc Reviewer is about to issue a verdict on a document change (status update, digest, decision record, report, knowledge entry), or Executor Reviewer is about to issue a verdict on an executed task (code changes, file edits, config updates, script runs). Covers the per-dimension rubric, the sp…
Orient
Use when the user wants a project-wide status briefing, asks what to work on next, or when autonomous scheduling needs a fresh situational assessment
Critique
Adversarial review of plans, findings, or designs before committing. Use before: multi-file plans, experiment conclusions (experiments/EXP-NNN.md) that others will cite, or architecture changes from /solution-design. Prioritize when irreversible or consumed by fleet workers.
Refresh Skills
Use when skills may be out of date with CORTEX.md conventions, after significant system changes, or for periodic skill health audits
Client Manage
Use when creating, deploying, updating, or troubleshooting cortex-client instances on remote devices
Gravity
Use when a manual fix or workaround keeps recurring and might need to be formalized into a convention, skill, or code
Approval
Use when the user wants to review, approve, or reject queued operations
Deep Retrospective
Use when reviewing multi-day work across past Claude Code sessions, extracting reusable patterns from historical logs, or asking what was learned about a topic over time.
Compound Simple
Use when completing a task and need lightweight post-task reflection
Develop
ALWAYS use this skill when implementing new features, fixing bugs, or modifying code in the Cortex codebase. Enforces test-driven development discipline: write tests first, then implement, then verify. Covers agent-server code, training scripts, data pipelines, and any code that will run unattended. Do NOT skip this skill and write code directly.
Compound
Use at the end of a work session, or when accumulated findings need to be embedded into conventions, skills, or patterns
Manager Method
Use when a Manager owns a composite task node and must (Phase A) decompose it into children or (Phase B) verify, accept, or rework a finished/blocked child. Covers judging decomposability, mapping seams before cutting, the Cut-at-the-Seam decomposition rules, template selection by residual reasoning, the per-child self-audit gate, acceptance-before-trust verification (incl. independent verifier c…
Thread
Use when working with the Cortex thread system — understanding thread architecture, writing or modifying thread config (the per-entity JSON files under config/thread-templates/ — agents, templates, shells; plus transitions and hooks fields within templates), debugging thread execution, or when the user asks about multi-agent pipelines, agent orchestration, or the !thread command. Also trigger whe…
Evolve
Use when Cortex needs a self-audit of its capabilities, when skill gaps are suspected, or when the last evolve was >7 days ago
Debug Campaign
Use when debugging multi-round issues on pipelines, services, or complex systems, when a fix-test cycle has gone through 3+ iterations without convergence, or when encountering bugs that resist simple fixes