AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Context Forge

skill-glowelephant-context-forge-context-forge · by glowElephant

|

No reviews yet
0 installs
39 views
0.0% view→install

Install

$ agentstack add skill-glowelephant-context-forge-context-forge

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets Used
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-glowelephant-context-forge-context-forge)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
2mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Context Forge? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

context-forge

Auto-harness-engineering for AI coding agents. Turns a 5-minute project discussion into a fully context-engineered GitHub repository.

Announce at start: "Using context-forge to bootstrap a harness-engineered repo for your project." (Translate to the user's language — see "Language matching" below.)

Language matching

Detect the user's language from their first message and respond in that language for ALL user-facing output (announce message, fixed questions, free discussion, Phase 3 proposal, Phase 4 prompts, Phase 6 hand-off). Code blocks, file paths, command names, and frontmatter keys stay in English. Catalog entry names (e.g., gstack-claude-toolkit) stay as-is.

If the user switches language mid-flow, switch with them. If unclear, ask once.

The example output blocks in this skill are written in English; translate them when speaking to the user.

Hard prerequisites

Before doing anything else, verify ALL of these. If any fail, ask the user to fix and stop.

  1. gh CLI is authenticated:

``bash gh auth status `` Expected: a logged-in account.

  1. context-forge is checked out locally with the catalog present. The user should set CONTEXT_FORGE_PATH to their clone (default: C:/Git/context-forge or ~/code/context-forge):

``bash test -d "$CONTEXT_FORGE_PATH/catalog" && test -f "$CONTEXT_FORGE_PATH/sources/index.json" ` If CONTEXTFORGEPATH` is not set, ask the user where their context-forge clone is and persist it for the session.

  1. The catalog passes validation:

``bash "$CONTEXT_FORGE_PATH/scripts/validate-catalog.sh" ``

If any step fails: stop and report the exact command that failed. Do NOT proceed.


Phase 1: Fixed questions (always ask, in order)

Ask these one at a time. Wait for each answer before asking the next.

  1. Project type? Multiple choice (single answer):
  • web (frontend / backend / fullstack)
  • game-engine (Unity, Unreal, Godot)
  • mobile (iOS, Android, cross-platform)
  • cli
  • library
  • mcp-server
  • other (free-text)
  1. Primary language / framework? Free text (e.g., "Next.js + TypeScript", "Unity 6 / C#").
  1. Solo or team? Multiple choice:
  • solo
  • small team (2–5)
  • larger team (6+)
  1. Multi-agent setup needed? (Claude Code subagents, multi-IDE coordination, etc.) yes / no.

After all 4 are answered, echo the answers back in a single short summary so the user can correct any.


Phase 2: Free discussion

After fixed questions, hold a focused conversation — one question per turn. Goal: understand goals, constraints, success criteria, and any domain knowledge that should land in docs/spec.md.

What to extract (do NOT ask all at once — pull out as relevant)

  • Goal — what does "done" look like in 1 sentence?
  • Top 3 milestones — name each in 5–10 words.
  • Hard constraints — deadlines, regulatory, perf budgets, target devices.
  • Domain knowledge — vocabulary, prior incidents, who the users are.
  • What to avoid — past failed approaches, anti-patterns the team rejects.

Stop condition

Stop asking when you can write all five bullets above. If you've asked more than 8 questions and still can't, summarize what you have and ask the user "anything else essential before we move on?"

Output of Phase 2

Internally produce:

goal: ...
milestones: [..., ..., ...]
constraints: ...
domain: ...
avoid: ...

Do NOT show this YAML to the user yet. It feeds Phase 3.


Phase 3: Catalog match

Read every catalog/*/*.md file's frontmatter from $CONTEXT_FORGE_PATH/catalog/. For each entry, decide whether it applies based on:

  1. Domain match — at least one of entry.domain overlaps with the user's project type, OR domain: [general] is present.
  2. when_to_use semantic fit — does the entry's when_to_use text describe this user's project? Use judgment.
  3. Multi-agent gating — entries with category: multi-agent only apply if Phase 1 Q4 was "yes".
  4. Stack-specific gating — if entry has applies_to_files (e.g., [".cursorrules"]) and the user's stack doesn't use those files, skip it.

Group results by category. Within each category, pick at most 3 (sort by priority field, fall back to alphabetical).

Present the proposal

Show the user a single message structured like this:

Based on our discussion, I'll include these from the catalog:

CLAUDE.md (claude-md):
  - karpathy-single-claude-md — Single careful CLAUDE.md
  - gstack-claude-toolkit — Garry Tan 23-tool stack

Skills:
  - superpowers-skills-pattern — Skills framework
  - cursor-rules-pattern (you mentioned Cursor)

Spec-driven:
  - github-spec-kit

MCP:
  - mcp-server-bootstrap

Prompts:
  - prompt-engineering-foundations

Anything to remove or any category you want me to add?

Wait for user response. If they ask to remove items, drop them. If they ask for items in a category that has none matched, search the catalog directly for that category.


Phase 4: Collect repo info

Ask in one message (multiple-choice + free-text):

  • Repo name? (free text, kebab-case suggested)
  • Public or private? (default: private)
  • Local clone path? (ask explicitly — never guess. Prefer the directory the user is currently working in.)
  • GitHub Issues for milestones? (default: no — opt-in, since this is destructive in a brand-new repo)

Phase 5: Create repo and populate

Run these in order. After each command, confirm the expected output before proceeding to the next.

# 1. Create remote + clone
gh repo create "/" --private --clone --add-readme=false   # or --public
cd "/"

# 1b. Normalize local branch to `main` — Windows defaults to `master`, which
#     mismatches the remote's expected `main` and causes `git push` to fail with
#     "src refspec main does not match any". Idempotent on macOS/Linux.
git branch -M main

# 2. Make directory structure
mkdir -p docs/know-how .claude/skills .claude/agents .claude/rules

# 3. Copy chosen catalog entries
#    For each  in the user-confirmed list:
cp "$CONTEXT_FORGE_PATH/catalog//.md" \
   "docs/know-how/--.md"

Then synthesize these top-level files (do not just copy from catalog — these are written from scratch using Phase 2 inputs):

  • CLAUDE.md — 30–80 lines. Sections: project goal (Phase 2), high-level architecture (if known), agent guardrails (avoid list), references to docs/know-how/*.md.
  • AGENTS.md — Mirror of CLAUDE.md but tool-agnostic (no /skill references, no Claude Code specifics). Used by Cursor / Copilot / Codex.
  • README.md — 1-line tagline + Goal + Quick start placeholder + Status section.
  • docs/spec.md — Phase 2 YAML expanded into Markdown sections: Goal, Milestones, Constraints, Domain, Avoid.

Then add boilerplate (only if catalog/boilerplate/.md exists):

# Optional - copy project-type .gitignore etc. if present
cp "$CONTEXT_FORGE_PATH/catalog/boilerplate/.gitignore" .gitignore

Always-include minimal .gitignore:

.env
.env.local
node_modules/
.DS_Store
Thumbs.db
.vscode/
.idea/

Commit and push:

git add .
git commit -m "초기 하네스 셋업 (context-forge)"
git push -u origin main

Optional — set repo description and topics from Phase 2 goal:

gh repo edit / --description ""
# topics: derived from project type and key tech words

Phase 6: Hand off

Print a final message in this exact shape:

✅ Done. Your harness-engineered repo:

  https://github.com//
  local: /

Included from catalog ( entries):
  

Next steps:
  cd /
  claude       # start a fresh session in the new repo

Tip: Edit docs/spec.md as your project evolves; CLAUDE.md/AGENTS.md should stay short.

After printing, stop. Do not start working in the new repo from this session.


Error handling

  • gh repo create fails with name collision — ask user for an alternate name; do not silently rename.
  • cp fails because catalog file missing — list the missing files and stop. Either the catalog wasn't synced or the entry name was misspelled.
  • git push fails — show the error, do not retry. Common cause: branch protection or repo visibility mismatch.
  • User aborts mid-flow — leave the partially-created repo in place; do not auto-delete.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.