KnowLoop is an interview-coaching skill for technical interview preparation. It uses one learning loop for algorithms, Python, system design, fundamentals, SQL and data questions, behavioral questions, introductions, and resume projects:
understand → attempt → probe → repair → retry → save → resume
The coach adapts to the answers it receives and resumes from saved evidence.
Install the interview-coach skill and the optional $coach shortcut, then
start with a topic, a piece of material, or a course task:
$coach
$coach What should I learn today?
$interview-coach Practice Python with me.
$interview-coach Start a system design mock interview.
$interview-coach Review my last weak point.
Give the coach a subject, goal, language, or time limit when those details matter. You can begin with reading, code, an introduction, or a real project; you do not need to complete a course first.
| Mode | What happens |
|---|---|
| Learn | Explain the current material, then check understanding. |
| Practice | Let you answer first; probe, repair, and retry. |
| Mock | Ask one unseen interview question at a time, then debrief. |
| Review | Re-test a due skill with a fresh or changed-condition question. |
Use $interview-coach when you want this learning loop. Ordinary fact lookup,
code editing, and direct explanations can remain ordinary conversation. Mock
interviews start only when you ask for one. Coach-authored examples do not count
as independent learner evidence.
System design follows this progression:
Functional Requirements → Non-functional Requirements → Core Entities → Data Flow → API / System Interface → High Level Design → Deep Dives
Start a new design with requirements, then advance from the evidence in the current stage. Learn, Practice, Mock, and Review can all use the same path.
flowchart TD
A[Choose a topic, material, or course task] --> B{Choose a mode}
B -->|Learn| C[Explain and check understanding]
B -->|Practice| D[Attempt, probe, repair, and retry]
B -->|Mock| E[Answer unseen questions]
B -->|Review| F[Retest a due skill]
C --> G[Save a checkpoint]
D --> G
E --> G
F --> G
G --> H[Resume from the next gap]
H --> B
After each assessable answer or retry, the coach synchronously saves a checkpoint before moving to the next prompt, stage, mode, or wrap-up. Learners do not run storage commands, write JSON or SQLite, or maintain a dashboard.
Learner records stay only in a local SQLite database under a private directory outside this repository, installed skills, and other Git repositories. The public repository contains general instructions, course data, and synthetic examples. Records are not uploaded automatically.
The dashboard is opt-in. Only an explicit request opens a transient loopback
view at 127.0.0.1; it reads directly from SQLite, keeps display state in
memory, and closes after page load or a bounded timeout. Saving never creates a
learner HTML or JSON report.
For an explicitly consented current Practice, Mock, or Review answer, the
interview-coach skill also includes a portable, dependency-free Jev advisory.
It receives only the current prompt, answer, and language on standard input,
returns observable current-text signals and validated usage, and persists
nothing. It never grades mastery or changes the canonical Coach record. See
current-answer advisory
for the explicit-consent, own-key, and local-only boundaries.
Requirements: Python 3.10 or newer for local storage. No database server, Node.js installation, or voice service is required.
From a checkout of this repository, copy the two skills into your Codex skills directory:
mkdir -p ~/.codex/skills
cp -R interview-coach coach ~/.codex/skills/If CODEX_HOME is configured, use its skills/ directory instead. Then start
$coach or $interview-coach in Codex.
The publishable output is limited to the local docs/ catalog,
docs/courses.json, and the explicitly synthetic docs/dashboard-demo.html.
These pages contain public course and demonstration data only; private learner
records remain in local SQLite.