codecall
Quiz you on the code your AI agent just wrote, so you actually understand the change.
- Source repo
- GAURAV-1313/codecall
- Stars
- ★ 16
- Last updated
- 7d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 55/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer who just had Claude Code add authentication middleware wants to confirm they understand the new security boundary, not just watch tests pass.
- Not a fit if
- Developers who only want more generated code and no comprehension checks
- Teams whose working tree is rarely clean enough for Git-based drills
- Source review
- 55/100 · Major gaps
What does this agent do, and when should you use it?
codecall is a skill for AI coding agents that turns each meaningful change into a short learning session built from your own code. It offers three modes: Learn (a ~4-minute quiz on the 2–4 ideas behind the latest change), Trace (walk one request from entry point to response, file:line by file:line), and Drill (a debugging round with one planted bug in your own project). The skill works inside Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Kiro and Copilot, and is installed via the Skills CLI, npm, or a Claude Code plugin. It requires no API key, account, or network call; the only files it writes are self-contained HTML receipts in ./codecall/. Drills temporarily edit one file and restore it with Git, without committing, stashing, or branching.
Learn mode is triggered by /codecall: it inspects the change the agent just made, announces a learning opportunity with an estimated time, waits for Start Learning / Skip, then adapts to your self-rated confidence (Expert / Comfortable / Heard of it / Never learned). It explains each idea using your code and runs two checks per idea; a correct answer makes the next check harder, a wrong one names the gap, re-teaches from your code, and asks a fresh question. Trace mode (/codecall trace POST /login) prints the request path as file:line steps, then checks you on the 2–3 steps most likely to hide a bug, and never edits code. Drill mode (/codecall drill src/orders medium) plants one realistic bug — an off-by-one, a missing await, an unhandled empty case, a wrong status code — gives only the symptom and repro, offers escalating hints (where to look, what to log, which function, which line), and on completion shows the planted change and restores the file with Git. A built-in map of about 80 backend concepts keeps terminology consistent and, when you miss a mapped concept, adds a single Go deeper line linking a free video chapter plus official docs (MDN, OWASP, PostgreSQL). Every session ends with a self-contained HTML receipt in ./codecall/ recording the change in one line, per-idea pass/fail marks, and a score of the form Could explain N of M.
- A developer who just had Claude Code add authentication middleware wants to confirm they understand the new security boundary, not just watch tests pass.
- An interview candidate preparing for debugging rounds who wants realistic practice with hints and a review on a project they already know.
- An engineer onboarding onto an unfamiliar codebase who runs a trace to learn the call chain from request entry to response.
- A tech lead who wants an auditable artifact showing a teammate could explain AI-generated changes before they ship.
- A backend learner who hits a gap in caching, queues, or webhooks and follows the receipt's Go deeper link to fill it.
How do you install or deploy this agent?
Pick one method, then restart your agent or start a new session:
npx skills add GAURAV-1313/codecallOr install globally via npm (works for Claude Code and Codex):
npm install -g codecallOr use the Claude Code plugin marketplace — the only option that automatically offers sessions after meaningful changes:
/plugin marketplace add https://github.com/GAURAV-1313/codecall
/plugin install codecall@codecallCodex users add the plugin instead:
codex plugin marketplace add https://github.com/GAURAV-1313/codecall
codex plugin add codecall@codecallHow do you use this agent?
The command depends on how you installed it: /codecall for Skills CLI or npm, /codecall:codecall for the Claude Code plugin, and $codecall for Codex.
Quiz yourself on what was just built:
/codecallFollow a single flow through your code:
/codecall trace POST /loginRun a debugging drill at a chosen difficulty (requires Git and a clean working tree):
/codecall drill src/orders mediumReceipts land in ./codecall/ inside your project; delete the folder to forget every session.
What are this agent's strengths and limitations?
- Every question is grounded in your own code, so receipts map directly to the change you just made rather than generic exercises.
- Adaptive retry: a wrong answer triggers a re-teach, a named gap, and a fresh question instead of ending the session.
- Runs entirely locally with no API key, account, or network call; receipts are the only files written.
- Drills touch a single file and restore it with Git, with no commit, stash, or branch — a tight, predictable blast radius.
- Trace mode emits concrete
file:linesteps, which is useful for mapping an unfamiliar call chain quickly.
- Drills refuse to start without Git and a clean working tree, so you must commit or stash first.
- The automatic post-change offer exists only in the Claude Code plugin; other installs require typing the command yourself.
- The backend concept map covers about 80 server-side concepts; changes outside it get no Go deeper line and no cross-session reinforcement.
- Receipts are standalone HTML files with no dashboard or team-level aggregation, so collecting results is manual.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| codecall This agent | 55 · Major gaps | Agent plugin / skillFree | ★ 16 | 7d ago | TypeScript | Codex · Claude Code |
| Waza | 63 · Some gaps | CLIFree | ★ 7.5k | 3d ago | Python | Codex · Claude Code |
| Self-Learning Skills | 56 · Major gaps | Agent plugin / skillFree + model costs | ★ 973 | 1mo ago | — | Codex · Claude Code |
| Loop Library: Reusable AI Agent Workflows | 61 · Some gaps | Agent plugin / skillFree | ★ 3.2k | 1mo ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
README and SECURITY.md state no network, no API key, writes only ./codecall/ receipts, drill edits one file and restores via git checkout, plugin hook only returns a text reminder; these align with package.json postinstall installer and bin entry, showing reasonable least-privilege and user-confirmation (Start Learning / Start Drill) design. However, sensitive-data handling is only a 'no collection' claim; it does not state whether receipt HTML embeds source snippets or paths, nor any redaction policy. Dependency security is weak: devDependencies all use 'latest' and postinstall runs a script on global install, creating supply-chain and reproducibility risk. External effects and rollback have a Git restore mechanism, but drill requires a clean working tree and cannot roll back uncommitted changes; README also advises approving the planted edit without reading it, weakening user confirmation. Deductions: insufficient sensitive-data and dependency-security evidence; user confirmation weakened in drill scenario.
README is consistent on commands, install methods, and troubleshooting (restart, command not found, drill needs clean tree); SECURITY.md and README corroborate network, file-write, and drill behavior, giving good self-consistency. But dependency availability is weak: devDependencies use 'latest', and postinstall depends on dist/install-skill.js existing at install time, which fails if build artifacts are missing; test/cli.test.ts invokes dist/cli.js without a prior build, an environment dependency risk. Failure messages: README Troubleshooting covers common failures, but code-level error messages are not shown, so evidence is limited.
README explicitly supports Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Kiro, and Copilot, with per-install command differences; scenarios cover Learn/Trace/Drill, and the audience is clear. Capability boundaries are stated: trace never edits code, drill edits one file, plugin hook only adds a reminder. But trigger precision is thin: the automatic offer relies on the plugin hook, and README does not explain hook trigger conditions or false-trigger handling; the 'approve without reading' advice may reduce trigger safety. Environment fit requires Node >=20 and a clean Git tree, limiting non-Git or uncommitted projects.
README has clear structure: Features table, Install table, command mapping, Learn/Trace/Drill examples, privacy, Troubleshooting, Contributing, License — good information architecture. Install notes cover multiple methods with restart guidance; naming is stable (codecall); examples and FAQ are adequate. Known limitations are only scattered (drill needs Git and clean tree) and not consolidated; versioning is only the 1.5.x support table in SECURITY.md, with no CHANGELOG; maintenance responsibility is unclear because the publisher is unverified, though SECURITY.md gives a reporting channel and 7-day reply promise. License is MIT and complete.
Output usability: receipts are self-contained HTML with change summary, concept mastery marks, score, and key points, directly usable. Marginal value: turns AI-generated code into learning and debugging practice, a clear gain for the target user. But cost-benefit: each session takes about 4 minutes, generates HTML files, and drill requires Git operations; for high-frequency changes this may cause file accumulation and time cost, and the automatic offer via plugin hook may add noise. Evidence does not show actual receipt template content quality, so deducted.
README claims (no network, only receipts, drill restore) are corroborated by SECURITY.md, but both are project self-descriptions with no third-party or code-level evidence. Tests cover backend-map validation, CLI help, and installer, but not network calls, file-write scope, or drill restore — key claims — so claim_traceability is limited. Cross-source corroboration is only within-repo documentation, with no independent external source. Fact/inference separation is weak: README mixes 'no network' claims with feature descriptions without distinguishing verified facts from design intent, so deducted.
- README advises approving the planted drill edit without reading it, which may weaken user confirmation awareness; users should still review the change.
- devDependencies all use 'latest' and postinstall runs a script on global install, creating supply-chain and reproducibility risk; pin versions and review the install script.
- Receipt HTML may embed source snippets or paths; README does not describe redaction, so inspect ./codecall/ contents before use on sensitive projects.
- Drill requires a clean Git working tree; uncommitted changes cannot be restored via git checkout, so commit or stash first.
- Publisher identity is unverified and maintenance responsibility/update path is unclear; monitor repository activity and security advisories for long-term use.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v1.3.0 codecall v1.3.0
Release notes excerpt
Every session now ends in a one-page receipt (./codecall/NNNN-<slug>.html) with the score "Could explain N of M" and a streak line. SKILL.md trimmed and re-led with the slot; one-line install: npx skills add GAURAV-1313/codecall. Test pins fixed.View on GitHub ↗ -
v1.2.2 codecall v1.2.2
Release notes excerpt
Small patch release - README now links the [landing page](https://codecall-theta.vercel.app/), npm, and GitHub at the top. - Lockfile refresh (no dependency changes, just pruned stale optional-platform variants). No behavior changes for any install path.View on GitHub ↗ -
v1.2.1 codecall v1.2.1
Release notes excerpt
Small patch release - README: added a second worked example (retry/backoff logic) alongside the existing auth example, so it's clear codecall isn't just for security-flavored code. - Fixed a high-severity dev-dependency advisory (nanoid, GHSA-2v37-7h3g-55p8) — dev-only, never shipped to users, but fixed for repo hygiene. - Removed a stray self-referential codecall dependency that had crept into package.json/package-lock.json from a local install in this directory — never actually published, but worth cleaning up. No behavior changes for any install path.View on GitHub ↗