Context Loader
Tiered, size-guarded context load that skips completed tracks and files over 1MB
Test report
- Verdict
- Works with setup
- Score
- Tested
- Jul 21, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · f924d54
Cloned the repo (GitHub API was rate-limited, git clone worked), copied skills/context-loader into a throwaway HOME and confirmed frontmatter parses to exactly name + description; the body references no scripts or bundled files, only inline ls/git commands, and I found no curl|sh, base64 or credential access. Ran the protocol literally twice against two real targets. On an ordinary Astro repo it degraded: no CLAUDE.md and no conductor/ dir meant 5 of 6 workflow steps and 2 of 4 tiers resolved to nothing, so the strict run read one file (package.json, 465 B) and emitted a summary with five [unknown] fields, while my no-skill baseline read README + package.json and produced an actual product/stack/route/open-items picture — README.md appears in no tier of the priority table, which is the core gap. On a Conductor-shaped fixture the skill won concretely: 7 files opened vs baseline's 9, it excluded the completed t-009-auth plan.md that baseline wasted a read on, its mandatory pre-read ls -la caught src/generated-schema.json at 1,200,008 bytes and logged "skipped: too large" where baseline had no such guard, and its output carried "Current Track: t-014-billing / Pending Tasks: 3 / Completed Tasks: 2" against baseline's unstructured prose. Trigger phrasings judged: "Load the project context before we start on this track" (load, correct), "Picking this repo back up — get me up to speed on the current track without burning context" (load, correct), "Run /orchestrator-supaconductor:implement on t-014-billing" (load, correct), "Explain what src/lib/data.ts does" (skip, correct), "Implement a dark mode toggle on the settings page" (I would load it, because the description literally says "Use when... implementing features" — false positive, 4/5). Verdict setup, not pass: installation is trivial but the named manual step is scaffolding conductor/ (or installing the parent plugin) — without it the skill produces a worse context load than doing nothing. Artifacts at /private/tmp/claude-501/-Users-afin-Desktop-Skill-Agregator/94648f54-e301-46fc-afd8-b98858a51ecf/scratchpad/ctxload/{A-baseline,B-skill,C-fixture-baseline,D-fixture-skill}.md
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 4/5
- Output vs. baseline 5/10
- Docs & honesty 4/5
What Context Loader does
Defines a token-budgeted procedure for loading project context at the start of a session: honor .claudeignore/.gitignore, enumerate files with git ls-files, read in a four-tier priority order (manifests, then Conductor product/tech-stack docs, then the active track spec and plan), check every file's size with ls -la before opening it, and cap the load at 15 files. Files over 500KB are read head-and-tail only, files over 1MB are logged as skipped, and plans belonging to completed tracks are never loaded. Triggers on "load context", "project context", or starting work on a track in the orchestrator-supaconductor workflow.
How to install Context Loader
git clone --depth 1 https://github.com/Ibrahim-3d/orchestrator-supaconductor.git /tmp/context-loader-src
mkdir -p ~/.claude/skills
cp -R /tmp/context-loader-src/skills/context-loader ~/.claude/skills/context-loader
# Single file, no scripts, no deps. Verified: SKILL.md lands at ~/.claude/skills/context-loader/SKILL.md
#
# IMPORTANT — this skill assumes a Conductor-scaffolded project. Tiers 2-3 and 5 of its 6
# workflow steps read conductor/product.md, conductor/tech-stack.md, conductor/tracks.md and
# conductor/tracks/<id>/{spec,plan}.md. On a repo without a conductor/ directory those all
# no-op and the load is thinner than just reading the README.
#
# For the full workflow that creates conductor/, install the whole plugin instead:
# /plugin marketplace add Ibrahim-3d/orchestrator-supaconductor
# /plugin install orchestrator-supaconductor
# then drive it with /orchestrator-supaconductor:go "<what you want built>"
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Context Loader
-
/context-loaderTiered, size-guarded context load that skips completed tracks and files over 1MB
It also activates on plain-language prompts like these:
-
Load context for this track before I start -
Give me project context without burning tokens -
Prepare context for implementing this feature
Frequently asked questions
- Is the Context Loader skill free?
- Yes. The skill itself is free from Ibrahim-3d/orchestrator-supaconductor. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Context Loader work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 21, 2026. Verdict: Works with setup. Cloned the repo (GitHub API was rate-limited, git clone worked), copied skills/context-loader into a throwaway HOME and confirmed frontmatter parses to exactly name + description; the body references no scripts or bundled files, only inline ls/git commands, and I found no curl|sh, base64 or credential access. Ran the protocol literally twice against two real targets. On an ordinary Astro repo it degraded: no CLAUDE.md and no conductor/ dir meant 5 of 6 workflow steps and 2 of 4 tiers resolved to nothing, so the strict run read one file (package.json, 465 B) and emitted a summary with five [unknown] fields, while my no-skill baseline read README + package.json and produced an actual product/stack/route/open-items picture — README.md appears in no tier of the priority table, which is the core gap. On a Conductor-shaped fixture the skill won concretely: 7 files opened vs baseline's 9, it excluded the completed t-009-auth plan.md that baseline wasted a read on, its mandatory pre-read ls -la caught src/generated-schema.json at 1,200,008 bytes and logged "skipped: too large" where baseline had no such guard, and its output carried "Current Track: t-014-billing / Pending Tasks: 3 / Completed Tasks: 2" against baseline's unstructured prose. Trigger phrasings judged: "Load the project context before we start on this track" (load, correct), "Picking this repo back up — get me up to speed on the current track without burning context" (load, correct), "Run /orchestrator-supaconductor:implement on t-014-billing" (load, correct), "Explain what src/lib/data.ts does" (skip, correct), "Implement a dark mode toggle on the settings page" (I would load it, because the description literally says "Use when... implementing features" — false positive, 4/5). Verdict setup, not pass: installation is trivial but the named manual step is scaffolding conductor/ (or installing the parent plugin) — without it the skill produces a worse context load than doing nothing. Artifacts at /private/tmp/claude-501/-Users-afin-Desktop-Skill-Agregator/94648f54-e301-46fc-afd8-b98858a51ecf/scratchpad/ctxload/{A-baseline,B-skill,C-fixture-baseline,D-fixture-skill}.md
- What is the Context Loader SkillProof Score?
- 7.2/10 — installs cleanly 5/5, triggers reliably 4/5, output vs. baseline 5/10, docs & honesty 4/5.
- How do I install Context Loader?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Context Loader with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.