Agent skill

Redlamp Bench

by pdcgomes in pdcgomes/redlamp

Redlamp Bench, the way an agent asks the owner to do a step in another app and gets the results back without anyone moving files.

MPL-2.0Auto-check passed

Install Redlamp Bench

skills CLI
$ npx skills add pdcgomes/redlamp --skill redlamp-bench -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install pdcgomes/redlamp redlamp-bench --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/pdcgomes/redlamp.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.cursor/skills/redlamp-bench .claude/skills/redlamp-bench && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
redlamp-bench
GitHub stars
222
Token cost
~2.1k tokens
SKILL.md length
1,153 words
Files
3
Skills in repo
11
Repo updated
First seen
Licence
MPL-2.0

At a glance

Redlamp Bench, the way an agent asks the owner to do a step in another app and gets the results back without anyone moving files.

  • Works in 4 steps: You run redlamp task new. The task lands… → When the owner's iPhone can reach the… → He follows the steps: share the photos… → …
  • You need the owner to check
  • SKILL.md covers How a task travels, Make the task, Make Lightroom's results… and Tell the owner, plus 2 more sections
  • Runs Swift scripts from its folder; calls mise

What it does

Redlamp Bench is an agent skill from pdcgomes/redlamp. Redlamp Bench, the way an agent asks the owner to do a step in another app and gets the results back without anyone moving files. The agent writes a bench task (a folder with a manifest, one-screen steps and reference images) with redlamp task new; the owner's iPhone pulls it from the Recipe Lab's hub, he follows the steps in Lightroom (or Prequel, Photos, another app) and shares the exports back, and the phone sends the task to Done once every result is paired. Use whenever you need the owner to check, measure…

Its SKILL.md is about 2.1k tokens, which your agent loads only when the skill is triggered. The skill folder holds 4 other files (for example `examples/dec-08/draft.json`).

The repository describes itself as: A native, open-source RAW photo editor for Mac, iPad and iPhone that feels immediately familiar to Lightroom users. The licence is MPL-2.0.

When your agent uses it

  • You need the owner to check
  • Produce something in Lightroom
  • Another app (Lightrooms masks
  • Exports against Redlamps)

Example prompts

  • “s iPhone pulls it from the Recipe Lab”
  • “s masks, sliders, presets or exports against Redlamp”
  • “s steps, or when reading a task”
  • “/redlamp-bench”

Workflow steps

4 steps, taken from the first numbered list in SKILL.md.

  1. You run redlamp task new. The task lands in ~/Library/Application Support/Redlamp/Bench/Outbox//, outside every checkout, so it doesn't…
  2. When the owner's iPhone can reach the Recipe Lab's hub (the harness open on the Recipe Lab with its Bench tab's Receive on), Redlamp Bench…
  3. He follows the steps: share the photos to Lightroom, do exactly what each step says, export, and share the exports back to Redlamp Bench…
  4. Once every photo has its result and every required question has an answer, the phone sends the task back. It's checked as untrusted input…

What it can do on your machine

Read from SKILL.md and the folder at commit bc52683. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships script files (Swift), which the agent can run.

    Shell commands in SKILL.md call:

    • mise

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Redlamp Bench loads about 2.1k tokens when it runs. Until then it costs about 201 tokens; SKILL.md has 1,153 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~201
When it runs · the whole SKILL.md, loaded when a task matches
~2.1k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from pdcgomes/redlamp at commit bc52683, republished under its MPL-2.0 licence (© pdcgomes). 1,153 words, ~2,137 tokens.

Download SKILL.mdSave it as .claude/skills/redlamp-bench/SKILL.md (or your agent's skills folder). This skill also uses 2 other files; get the full folder from GitHub.
name
redlamp-bench
description
Redlamp Bench, the way an agent asks the owner to do a step in another app and gets the results back without anyone moving files. The agent writes a bench task (a folder with a manifest, one-screen steps and reference images) with `redlamp task new`; the owner's iPhone pulls it from the Recipe Lab's hub, he follows the steps in Lightroom (or Prequel, Photos, another app) and shares the exports back, and the phone sends the task to Done once every result is paired. Use whenever you need the owner to check, measure, compare or produce something in Lightroom or another app (Lightroom's masks, sliders, presets or exports against Redlamp's), when you're about to ask him to add photos to Lightroom and send results back, when writing such a task's steps, or when reading a task's results.

Redlamp Bench: a step in another app, as a task

When work needs Lightroom's own answer, such as how its masks combine, what a slider does to a ramp or what a preset renders, don't ask the owner to move photos around. Make a bench task. It reaches his phone by itself, opens on its steps one at a time, and comes back to a folder you can read, with each result paired with the photo it came from. docs/bench-tasks.md has the format, the hub and the app in full.

Use it for: anything only the owner can do in another app, with files going in and coming out: Lightroom checks (masks, sliders, profiles, presets, exports), and a phone app's output on given images.

Don't use it for:

  • What you can measure yourself, from the CLI or tests.
  • A question with no files: put that on your workstream canvas's Needs you instead.
  • A phone app's filter to rebuild as a Redlamp look: the owner makes those himself on the phone as look references (New Look).

How a task travels

  1. You run redlamp task new. The task lands in ~/Library/Application Support/Redlamp/Bench/Outbox/<id>/, outside every checkout, so it doesn't matter which worktree you're in.
  2. When the owner's iPhone can reach the Recipe Lab's hub (the harness open on the Recipe Lab with its Bench tab's Receive on), Redlamp Bench pulls the task.
  3. He follows the steps: share the photos to Lightroom, do exactly what each step says, export, and share the exports back to Redlamp Bench. Each export pairs with its photo.
  4. Once every photo has its result and every required question has an answer, the phone sends the task back. It's checked as untrusted input and filed in .../Bench/Done/<id>/, and it leaves the outbox.

Make the task

Build the CLI once (SCHEME=redlamp mise run build), then use build/DerivedData/Build/Products/Debug/redlamp. It's written redlamp below.

Quick form, for a few photos and plain steps:

bash
redlamp task new --title "Sky masks on the coverage set" --app "Lightroom" \
    --workstream masking --tracker MSK-16 --issue 52 \
    --step "Select Sky on each | Masking › Create New Mask › Select Sky, nothing else." \
    --step "Show the mask | Set the mask's Exposure to −2.00, nothing else." \
    --step "Export | JPG, quality 100, full size, sRGB, no output sharpening, original file names." \
    --question "edge | Did Select Sky miss any sky at the trees? | Yes, No, Partly" \
    photos/*.dng

Without --no-auto-steps, it adds a first step that shares the photos to the app and a last one that waits for the results. Add --manual when the number of results isn't one per photo; the owner then marks the task done.

Draft form, when steps need actions, pictures or their own pairing: write a draft as task.json's fields, with each asset as {"source": "path", "id", "label", "counts"} and step pictures listed in "pictures" (copied into pictures/, named in steps as pictures/<name>). Then run redlamp task new --draft draft.json. examples/dec-08 is a complete one, with the script that made its photos.

Then run redlamp task check <id>, and fix every error and warning.

Steps that work on a phone

The app shows one step at a time, full screen, while the owner switches between it and Lightroom:

  • One action per step: "Frame 1: gradient A", not "make the gradients and export". check warns when a step's detail won't fit one screen (over 320 characters) or its title is over 60.
  • Lightroom's own words and exact values: menu paths as Lightroom shows them (Masking › Create New Mask › Linear Gradient), every value with its sign and decimals (Exposure −2.00), and "nothing else" when other settings must stay at their defaults.
  • Say how to export, once, in its own step: TIF if offered (otherwise JPG at quality 100), full size, sRGB, no output sharpening, and original file names.
  • Give a step an action when it has one. share and save put the photos one tap away; results waits for the exports and ticks itself; answer asks a question.
  • A picture ("picture": "pictures/where.png") when a setting is hard to find.
Assets that pair
  • Name assets for what they're for (1-gradient-a.tif, exposure-plus-50.jpg): with original file names on export, each result pairs by name.
  • Make look-alike assets distinguishable. When several assets are the same picture with different instructions, put a visible label in a corner (as the DEC-08 frames do). That tells the owner which is which in Lightroom, and lets similarity pair them if Lightroom renames the exports.
  • Keep tasks small: at most 20 assets and 1 GB.
  • Use fixture, CC0 or synthetic images only, never the owner's own photos.
Show full SKILL.md (466 more words)Show less

Make Lightroom's results measurable

Lightroom only gives back rendered images, so design the task so that the export answers the question:

  • A mask: apply one fixed adjustment inside it (Exposure −2.00, nothing else) on a flat or known image, and export an untouched copy as the baseline. In linear light, Exposure −2 scales by 2^(−2m), so the mask's strength at each pixel is m = −log2(out / baseline) / 2. Keep the adjustment moderate: −4 crushes the strong end into a few 8-bit levels.
  • Combining masks: two masks that vary along different axes (two crossing linear gradients), exported alone and combined. A product and a minimum differ everywhere both are partial. This is DEC-08's task, examples/dec-08.
  • A slider: one copy of the photo per value (exposure-minus-100, …, exposure-plus-100), each with its own step, so each result pairs by name and its value is known.
  • Profiles and presets: apply them to a known image, such as the capture kit's charts (build/app-looks/kit) or a fixture. For a phone app's filter to rebuild as a look, see the owner's look references instead.

Results are references only: never commit them or ship them, and never convert or ship Adobe's files (AGENTS.md).

Tell the owner

Add a Needs you item on your workstream canvas (.cursor/skills/workstream-canvas/SKILL.md):

  • Title: "Do the <task title> task in Lightroom".
  • Detail: where it comes from ("It appears in Redlamp Bench on your phone while the harness's Recipe Lab is open, Bench tab on"), roughly how long it takes, and what you'll do with the results.
  • Unblocks: the row it unblocks; mark it blocking when your work waits on it.
  • Command: mise run harness -- --scene recipe-lab --lab-tab bench.

Then carry on with other work; don't wait in a loop for hours.

Read the results

  • redlamp task show <id>: each asset's result, how it paired (file name, barcode, similarity with its score, or by hand), the answers and the owner's note. Add --json for the manifest and results.json together.
  • redlamp task wait <id> --timeout 60: returns once the task is in Done and complete; exit 2 means it isn't yet.
  • The files: in Done/<id>/results/, byte for byte as the other app exported them. results.json names each one's asset. An unpaired result (asset absent) is one the phone couldn't place: compare it with the assets yourself, or ask the owner.
  • Record what you measured: on your canvas, and in the tracker row's Status or the decision it settles. Name the task's ID.

Changing or taking back a task

  • Before the phone has it, edit the outbox folder freely, then run redlamp task check again.
  • After it's been pulled: run redlamp task withdraw <id>. The phone drops it unless it already has results. Then make a new task with a new ID. Never reuse an ID.
  • Done folders stay, as the record of what Lightroom answered.

© pdcgomes, MPL-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files in .cursor/skills/redlamp-bench of pdcgomes/redlamp.

  • SKILL.md
  • examples/dec-08/draft.json
  • examples/dec-08/make-grey.swift

Open the folder on GitHubat commit bc52683

Compare with similar skills

Redlamp Bench next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Redlamp Bench compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Redlamp Bench this skillpdcgomes/redlamp222—~2.1kAutomated safety check: PassMPL-2.0
Harness Benchruvnet/ruflo74k—~586Automated safety check: NotesMIT
Sandbox Benchvercel/next.js143k—~4.1kAutomated safety check: PassMIT
Bench Readgithub/awesome-copilot40k—~747Automated safety check: PassMIT
Benchddalcu/mlx-serve1.8k1 repos~1.1kAutomated safety check: PassCustom licence
Terminal Bench Looppaperclipai/paperclip99k—~6.3kAutomated safety check: PassMIT

Similar skills

  • Harness Bench

    ruvnet/ruflo

    Manage @metaharness/darwin bench suites — bench create {repo} scaffolds a JSON suite from a repo's test corpus; bench verify {suite.json} checks suite well-formedness.

    74k GitHub stars~586 tokensUpdated today
    DevelopmentAuto-check: notes
  • Sandbox Bench

    vercel/next.js

    Official

    Benchmark React or Next.js changes on Vercel Sandbox VMs with paired A/B statistics: react PR/commit vs base, or Next.js PR/commit vs base, measured end-to-end through the bench/render-pipeline app…

    143k GitHub stars~4.1k tokensUpdated today
    Data & AnalyticsAuto-check passed
  • Bench Read

    github/awesome-copilot

    Official

    Read artifacts from the shared bench — the workspace where desks leave findings, verdicts, and work products for each other and the operator.

    40k GitHub stars~747 tokensUpdated yesterday
    Auto-check passed
  • Bench

    ddalcu/mlx-serve

    mlx-serve benchmarking methodology — bench.sh/llmprobe usage, comparison-trap rules (same-methodology cells only, spec-decode variance, thermal lies, engine naming), perf-claim etiquette.

    1.8k GitHub starsUsed in 1 repo~1.1k tokens
    Media & CreativeAuto-check passed
  • Terminal Bench Loop

    paperclipai/paperclip

    Run one Terminal-Bench task through a bounded Paperclip smoke/diagnosis/fix loop.

    99k GitHub stars~6.3k tokensUpdated today
    Auto-check passed
  • Run @metaharness/darwin security bench (upstream "Darwin Shield" / ADR-155) — evolves a champion security-detection harness against a 10-vuln / 9-decoy corpus and grades it on…

    74k GitHub stars~1.1k tokensUpdated today
    DevelopmentAuto-check: notes

More from pdcgomes/redlamp

All 11 skills in this repo
  • Redlamp Manual

    pdcgomes/redlamp

    The Redlamp User Manual (docs/manual), a dense, book-style PDF for photographers, typeset from Markdown and a print style sheet by Chrome, with its ranges, defaults and shortcuts generated from the…

    222 GitHub stars~2.8k tokensUpdated today
    Auto-check passed
  • Redlamp Promo Studio

    pdcgomes/redlamp

    The promo studio, which makes Redlamp's short promotional videos for social feeds the way a small creative agency does - a brief, hooks and copy, a script on a music grid, visuals from the promo…

    222 GitHub stars~2k tokensUpdated today
    Auto-check passed
  • Redlamp Site

    pdcgomes/redlamp

    Map of redlamp.app, the Next.js website in web/: its stack and commands, its pages, the home page's sections with their anchors, components and content sources, the brand's Tailwind tokens and…

    222 GitHub stars~2.9k tokensUpdated today
    Auto-check passed
  • Redlamp Blog

    pdcgomes/redlamp

    The blog room, where every redlamp.app blog post is tracked from idea to announcement in a canvas the owner marks as they post.

    222 GitHub stars~2.1k tokensUpdated today
    Auto-check passed
  • Redlamp Press

    pdcgomes/redlamp

    The press room, where every outlet Redlamp is pitched to (newsletters, Mac and photography sites, developer and open-source communities, lists and directories) is tracked from finding it to its…

    222 GitHub stars~1.3k tokensUpdated today
    Auto-check passed
  • Redlamp Release

    pdcgomes/redlamp

    The release room, where every Redlamp release is prepared, checked, approved and shipped.

    222 GitHub stars~3.5k tokensUpdated today
    Auto-check passed

Questions about Redlamp Bench

What does Redlamp Bench do?

Redlamp Bench, the way an agent asks the owner to do a step in another app and gets the results back without anyone moving files. Redlamp Bench is an agent skill from pdcgomes/redlamp. Redlamp Bench, the way an agent asks the owner to do a step in another app and gets the results back without anyone moving files.

When should I use Redlamp Bench?

Redlamp Bench fits situations like: you need the owner to check; produce something in Lightroom; another app (Lightrooms masks; exports against Redlamps).

How do I install Redlamp Bench in Claude Code?

Run `npx skills add pdcgomes/redlamp --skill redlamp-bench -a claude-code`. Or copy the skill folder (.cursor/skills/redlamp-bench in pdcgomes/redlamp) into .claude/skills/redlamp-bench in your project. Claude Code loads it when a task matches its description.

How do I install Redlamp Bench in Codex?

Run `npx skills add pdcgomes/redlamp --skill redlamp-bench -a codex`. Or copy the skill folder (.cursor/skills/redlamp-bench in pdcgomes/redlamp) into .agents/skills/redlamp-bench in your project. Codex loads it when a task matches its description.

Can I use Redlamp Bench in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add pdcgomes/redlamp --skill redlamp-bench -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/redlamp-bench, .gemini/skills/redlamp-bench, .github/skills/redlamp-bench and .opencode/skills/redlamp-bench in your project.

What does Redlamp Bench need to run?

Going by SKILL.md and its folder, Redlamp Bench needs Swift for the scripts in its folder and the command-line tools its instructions call (mise).

Does Redlamp Bench access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Redlamp Bench safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Redlamp Bench use?

Redlamp Bench is published under the MPL-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Redlamp Bench use?

About 2.1k tokens (SKILL.md is roughly 8.5k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Redlamp Bench?

Skills that share tags, products or a category with Redlamp Bench: Harness Bench (ruvnet/ruflo, 74k stars), Sandbox Bench (vercel/next.js, 143k stars), Bench Read (github/awesome-copilot, 40k stars) and Bench (ddalcu/mlx-serve, 1.8k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Redlamp Bench?

pdcgomes (a GitHub user) maintains it in pdcgomes/redlamp, which has 222 GitHub stars. The repository holds 11 skills in this directory. The repository was last updated on October 10, 2026.

Source: pdcgomes/redlamp on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.