agentleFS
makukhin-iemakukhin-ie/ai-toolbox/skills/prusaslicer-filament-profiler/SKILL.md

prusaslicer-filament-profiler

Skill0 starsChanged 4 months agoMITView on GitHub

Generates ready-to-import PrusaSlicer filament + print profiles tuned to a specific material, brand, and printer. Use this skill whenever the user mentions PrusaSlicer settings, asks to create or fix a filament profile, reports print defects like stringing, under-extrusion, poor layer adhesion or warping with a specific filament, mentions a new spool of PLA/PETG/ABS/ASA/TPU/PC/PA, asks "what settings should I use for X filament on Y printer", or shares photos of a failed 3D print and wants a corrected profile. Trigger even if the user only mentions one half (just the printer, or just the material) — ask the other.

The file

Show its rules and text

We split this skill into the rules it gives an agent

18 rules. About 0 are common, 14 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈14

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Don't trigger when:

    line 27
  2. Distinctive:

    If the user has already given details earlier in the conversation (photos, screenshots, previous attempts, brand mentions), extract from …

    line 68
  3. Very specific:

    Read references/material-baselines.md for the material's baseline numbers.

    line 74
  4. Distinctive:

    Ask the user these (batch as needed, max 2-3 questions):

    line 82
  5. Distinctive:

    If the user reports a defect that fits the "Knocked-off small features," "Bed adhesion locally failing," or "Mid-print discoloration" …

    line 96
  6. Very specific:

    Use the template in references/ini-template.md.

    line 102
  7. Very specific:

    ALWAYS use this exact structure.

    line 116
  8. Distinctive:

    Never invents brand-specific data when uncertain.

    line 173
  9. Distinctive:

    Never recommends temperatures outside the safe range for the material (e.g.

    line 174
  10. Distinctive:

    Never disables the cooling fan entirely for materials that need it (PLA, PETG).

    line 175
  11. Very specific:

    Never produces a profile with inherits = left empty in [filament:...] or [print:...] sections — that's the bug that triggered the rewrite.

    line 176
  12. Distinctive:

    Never asks the user to manually edit dozens of fields in the slicer GUI when a complete .ini can be generated.

    line 177
  13. Distinctive:

    When the user shares photos of a print problem, resist the urge to confidently name a parameter to change.

    line 181
  14. Distinctive:

    Distinguish "geometry-localized" from "filament-wide.": If a defect appears only on certain features (tall bosses, small islands, sharp …

    line 189
  15. Distinctive:

    State diagnostic uncertainty out loud: If two or three causes could explain the photo, say so and ask the user one targeted question to …

    line 193
  16. Distinctive:

    If multiple diagnostic attempts in one session have been wrong, stop changing parameters and pivot to hardware/environment questions (Step …

    line 195
  17. Distinctive:

    Only after drying is confirmed do parameter changes get considered

    line 207
  18. Distinctive:

    Exception: if the user explicitly confirms drying was done recently (within last few prints) AND the symptoms are still present, then move …

    line 213

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PrusaSlicer Filament Profiler

What this skill produces

A single .ini file in PrusaSlicer's Import Config format (all parameters in one flat file, no [section] headers), containing both filament settings and print settings, tuned to the user's specific material + printer + observed problems. On import via File → Import → Import Config, PrusaSlicer auto-distributes parameters to the Filament Settings tab and Print Settings tab.

This format was chosen because:

  • [filament:...] / [print:...] section format with empty inherits = creates profiles that load as "default" with all parameters blank.
  • Config Bundle format has the same problem unless you explicitly inherit from an existing system profile, which varies per user installation.
  • The flat-no-sections format reliably works because PrusaSlicer applies parameters to the currently-selected profiles, which the user then saves under a new name.

When to use this skill

Trigger eagerly when any of these signals appear:

  • User mentions PrusaSlicer or Slic3r-based slicers and wants settings
  • User reports stringing, oozing, under-extrusion, gaps, layer separation, warping, poor adhesion, blobs, zits, ringing, or banding — with a specific material in mind
  • User has a new spool and wants a starting profile
  • User shares photos of a failed print and the cause appears material-related
  • User asks "what should I use for [material] on [printer]"
  • User mentions a specific filament brand (ROSA 3D, Prusament, Polymaker, eSun, SUNLU, Overture, Bambu, Hatchbox, etc.)

Don't trigger when:

  • User asks general 3D printing theory not tied to a profile
  • User asks about another slicer AND declines the manual-porting path (see «Other slicers» above)
  • User wants firmware tuning, hardware mods, or model design help

⚙️ Configure your printer fleet

references/printer-quirks.md ships with two fully-documented example printers that cover the two most common architectures:

  1. Anycubic Kobra 2 Pro — direct-drive, Klipper-based (representative of modern fast bedslingers)
  2. Anycubic Mega S — Bowden, old Marlin (representative of older budget printers)

Add your own printers to references/printer-quirks.md following the same structure (extruder type, firmware, retraction ranges, acceleration limits, known quirks). If your printer isn't listed, the skill still works: it will ask about extruder type (direct vs Bowden) and firmware, then apply generic adjustments for that architecture.

If the user mentions a printer by name, use that one. If they don't — ask; do NOT default silently. Direct-drive and Bowden printers need very different retraction, Linear Advance, and acceleration settings; getting it wrong produces an unusable profile. Phrase the question naturally — just the printer name, don't list quirks.

Other slicers (Cura, Orca, Bambu Studio)

This skill generates PrusaSlicer flat Import-Config .ini files. If you use another slicer:

  • Orca Slicer / Bambu Studio / SuperSlicer — parameter names largely overlap with PrusaSlicer; you can port values manually, but the .ini won't import directly. To add native support, create a new output template alongside references/ini-template.md (e.g. orca-template.md) with that slicer's JSON/ini structure and reference it in Step 3.
  • Cura — different parameter naming entirely; use the values from this skill's baselines (temps, retraction, speeds, accelerations) and enter them in Cura's UI manually, or add a cura-template.md mapping.

The material science in references/material-baselines.md and references/troubleshooting.md is slicer-agnostic — temperatures, drying, cooling and diagnosis apply everywhere.

Workflow

Step 1 — Gather inputs

Required dimensions. Ask only the ones you can't infer from context (max 3 questions, batch them).

DimensionHow to elicitDefault if user can't say
Material type"PLA, PETG, ABS, ASA, TPU, PC, PA?"Ask — never guess
Brand"Which brand?""Generic [material]"
Printer model"Which printer?"Ask — affects extruder type, accels
Nozzle sizeOnly ask if user mentions composites0.4 mm
Observed problems"What's going wrong?" or infer from photos"General starting profile"
Use caseOnly ask for unusual requests"Quality print, 0.20 mm"
Speed tierOnly ask if PLA/PETG and user mentions speed concernsBalanced

If the user has already given details earlier in the conversation (photos, screenshots, previous attempts, brand mentions), extract from there silently — don't re-ask.

Speed tier defaults: For PLA and PETG, the Balanced tier in material-baselines.md is the default. Only use Quality if the user is having quality issues, has tall thin features, or explicitly asks for quality-first. Only use Speed if the user explicitly asks for it AND has confirmed PA + Input Shaper are calibrated. Don't volunteer Speed tier preemptively — it requires a tuned printer.

Step 2 — Load baseline + apply overrides

Read references/material-baselines.md for the material's baseline numbers. Read references/printer-quirks.md if the printer is listed there. Read references/troubleshooting.md if the user reported specific defects — apply the override deltas from that file on top of the baseline.

The mental model is: baseline → printer adjustments → problem-specific overrides → user-specific overrides. Always apply in that order.

Step 2.5 — Hardware/environment check (CRITICAL — do not skip)

Before generating any profile changes for a user reporting defects on a previously-working setup, run through this checklist. Most reported defects in established print pipelines are caused by hardware/environment drift, not slicer parameters. Profile changes won't fix a knocked-off boss caused by a tilted printer — and trying to fix it with parameters wastes the user's time and filament.

Ask the user these (batch as needed, max 2-3 questions):

  1. "This setup worked before with another filament/profile?" — If yes, the slicer profile is probably not the root cause. Focus diagnosis on what changed: new spool, moved printer, ambient temperature, nozzle wear.
  2. "When did you last level the printer / check bed mesh?" — If "never" or "weeks ago," auto-level first before changing settings. Especially relevant after physically moving the printer.
  3. "Is the printer physically level (frame parallel to gravity)?" — Bedslingers (Kobra 2 Pro, Ender 3, etc.) are extra sensitive. A 1° tilt of the frame causes bed inertia to throw small features sideways during fast Y moves. Suggest checking with a spirit level on the printer's bottom frame.
  4. "When did you last clean the bed / nozzle?" — Burnt-on filament residue on the bed (especially PETG) creates local adhesion issues. Buildup on the outside of the nozzle drops chunks onto prints mid-job.
  5. "Did you purge the nozzle when switching filaments?" — Old filament left in the hotend mixes with new material, and may degrade (turn yellow/brown) if it sits at high temp for long. Especially visible on transparent/natural filaments.
  6. "Where do you store your filament — climate-controlled indoor space, or garage/basement/shed?" — Storage environment is the strongest predictor of moisture content, more than spool age. Garage/basement/shed storage means assume every spool is moisture-saturated regardless of packaging, due to temperature cycling. In that case, drying becomes mandatory before every print, not optional. See troubleshooting.md → "Filament storage" for details.

If the user reports a defect that fits the "Knocked-off small features," "Bed adhesion locally failing," or "Mid-print discoloration" patterns in troubleshooting.md, lead with hardware diagnosis, not parameter tuning. A bad profile rarely causes these; a tilted printer or dirty nozzle very often does.

When describing hardware procedures for a specific printer, web-search the model + component first. Belt tensioners, motor pulley access, bed leveling mechanism, control board specifics — these vary per model and the assistant's training data is unreliable at this level. For example: Kobra 2 Pro has adjustable belt tensioner knobs; Mega S has no belt tensioner at all (fixed clamp). Don't assume; verify.

Step 3 — Generate the .ini

Use the template in references/ini-template.md. Fill in every parameter — don't leave any blank, because PrusaSlicer "Import Config" replaces the currently-selected profile, so blanks become blanks.

Save the file to /mnt/user-data/outputs/ with a descriptive name: {Brand}_{Material}_{Printer}.ini, e.g. ROSA3D_PETG_Kobra2Pro.ini.

Step 4 — Present to the user

Present the file with present_files. Include in the conversation message:

  1. One-paragraph rationale — what was tuned and why, in plain language tied to the user's situation. Don't dump tables of every parameter; highlight the 3-5 most impactful changes.
  2. Import steps — short, numbered. Always remind the user: File → Import → Import Config (NOT Config Bundle), then save profiles under new names.
  3. First-action priority — if the user reported stringing or under-extrusion, the #1 action is almost always dry the filament. Lead with this. The profile is the second action, not the first.

Output format — the .ini file

ALWAYS use this exact structure. Header comments first, then parameters grouped by purpose with section comments. No [section] headers — keep it flat.

# generated by PrusaSlicer on YYYY-MM-DD
# Filament + Print settings: {Brand} {Material} - {Printer}
#
# HOW TO IMPORT:
#   1. Select an existing filament profile to overwrite (any user profile).
#   2. File -> Import -> Import Config (NOT Config Bundle).
#   3. PrusaSlicer applies filament params to Filament tab, print params to Print tab.
#   4. In each tab click the save icon and save as new profile name.

# ====================== FILAMENT SETTINGS ======================
{filament parameters}

# ====================== PRINT SETTINGS ======================

# --- Layers & perimeters ---
{layer params}

# --- Infill ---
{infill params}

# --- Skirt ---
{skirt params}

# --- Speeds (mm/s) ---
{speed params}

# --- Accelerations (mm/s^2) ---
{accel params}

# --- Slowdown for short layers ---
slowdown_below_layer_time = {value}

# --- Extrusion widths ---
{width params}

The full parameter list per material type is in references/ini-template.md. Always include all those parameters; do not omit any.

Note on perimeter_generator = arachne: This is the default in the template. Arachne is PrusaSlicer's variable-width perimeter generator — it dynamically adjusts extrusion width to handle thin walls, small features, and tight corners better than the legacy Classic generator. Especially important for parts with bosses, small holes, or organic shapes. There's almost no downside vs Classic except slightly slower slicing.

Filament notes field — drying and troubleshooting

The filament_notes field is a critical part of every profile. It MUST include:

  1. Material name and printer
  2. Drying recommendation (hours + temperature)
  3. Two short troubleshooting blocks: "If stringing remains" and "If under-extrusion remains" with concrete next steps

Pattern:

filament_notes = "{Brand} {Material} for {Printer}.\n*** Dry Xh at Y°C before first use. ***\n{Material} is hygroscopic - wet filament = strings + under-extrusion.\n\nIf stringing remains after drying:\n - increase retraction to X.X mm in 0.5mm steps\n - drop temp 5C at a time\n - print retraction tower\n\nIf under-extrusion remains after drying:\n - check for partial nozzle clog (cold-pull)\n - raise temp by 5C\n - check extruder tension\n - raise extrusion_multiplier"

What this skill never does

  • Never invents brand-specific data when uncertain. If a niche brand is mentioned, default to the generic baseline for that material and say so.
  • Never recommends temperatures outside the safe range for the material (e.g. PLA above 230°C, PETG above 260°C).
  • Never disables the cooling fan entirely for materials that need it (PLA, PETG).
  • Never produces a profile with inherits = left empty in [filament:...] or [print:...] sections — that's the bug that triggered the rewrite. Always use the flat no-sections format.
  • Never asks the user to manually edit dozens of fields in the slicer GUI when a complete .ini can be generated.

Diagnostic conservatism with photos

When the user shares photos of a print problem, resist the urge to confidently name a parameter to change. Photo-based diagnosis is harder than it looks, and getting it wrong costs the user a filament spool and printer hours. Specifically:

  1. Don't diagnose from a stopped print. A print stopped at layer 2 has no top surfaces, no overhangs, no tall features — what looks like a defect may just be how a first layer looks for that material. Ask if the print was allowed to complete or stopped early. If stopped, ask the user to let the next attempt run to at least the point where the suspected feature is fully printed.
  2. Distinguish "defect" from "normal-but-unfamiliar." Texture from the bed (PEI grain on the bottom), visible infill pattern on the first solid layer, surface haze on a fresh PETG print — these are not defects. Identify them as normal before suggesting changes.
  3. Distinguish "same defect on every filament" from "this filament only." If the user reports a defect that didn't happen with previous filaments using the same profile, it's the filament or calibration, not the profile parameters. Don't change the profile to compensate for one filament without first ruling out: wet spool, wrong material variant (transparent vs colored), bad batch, or calibration drift.
  4. Distinguish "geometry-localized" from "filament-wide." If a defect appears only on certain features (tall bosses, small islands, sharp corners) and not on equivalent flat areas, the cause is mechanical (Z-hop, accelerations, cooling on small features), not material. Don't recommend material changes.
  5. Distinguish "in the part" from "on the part." Brown/yellow blobs on the surface may be debris from the nozzle. Brown/yellow integrated into the material itself (visible through translucent prints, present in fresh extrusions) is thermal degradation. Different causes, different fixes.
  6. State diagnostic uncertainty out loud. If two or three causes could explain the photo, say so and ask the user one targeted question to disambiguate, instead of picking one confidently. Phrase: "This could be X or Y — could you check Z so I know which?"

If multiple diagnostic attempts in one session have been wrong, stop changing parameters and pivot to hardware/environment questions (Step 2.5 checklist). The pattern of "profile keeps not fixing it" is itself diagnostic — it usually means the cause is outside the profile.

Drying first, parameters second (CRITICAL RULE)

For PETG, TPU, PC, PA, Nylon, and ABS — when the user reports stringing, surface defects, brown/yellow discoloration in the material, weak layer adhesion, or unstable extrusion, drying must be confirmed as actively done before suggesting any parameter changes. Not "the spool is in a box with silica gel." Not "I dried it last month." Actively dried: a specific session of 3+ hours at the material's safe temperature, completed recently, with the print attempted within 24h of drying.

The reasoning: water in hygroscopic filaments produces symptoms that perfectly mimic profile problems (stringing, under-extrusion, surface defects, even local color changes). Profile changes will appear to "almost fix" the issue, leading to endless tweaking that never solves the root cause. Meanwhile a 3-hour dry would remove ~70% of the problem and reveal whether the profile actually needs adjustment.

Pattern that previously caused failure:

  • User reports defect → assistant suggests parameter X → user changes X → defect persists → assistant suggests parameter Y → defect persists → eventually user dries the filament → defect mostly resolves
  • The correct flow:
  • User reports defect → assistant asks "have you actively dried this spool, hours and temperature?" → if no, drying is the first recommendation, period
  • Only after drying is confirmed do parameter changes get considered

Phrasing this firmly: "Before I suggest any settings changes, I need to know: did you actively dry this spool, when, and at what temperature? Wet PETG/TPU/PA/PC produces symptoms that look exactly like settings problems but are not — and adjusting settings to compensate won't fix it. Even a partial dry (3-4 hours at the safe temperature) often resolves these issues entirely. So this matters before anything else."

Don't soften this to "you might want to try drying." Make it the precondition. The diplomatic version of "maybe try drying" has been demonstrated to fail repeatedly — users skip the recommendation because it's framed as optional. Frame it as gating: no parameter changes until dry status is confirmed.

Exception: if the user explicitly confirms drying was done recently (within last few prints) AND the symptoms are still present, then move on to other diagnoses. The rule is about defaulting to drying, not blocking all other help.

Verify hardware specifics with web search before stating them categorically

When making categorical statements about how a specific printer's hardware works — belt tensioners, motor pulleys, control boards, firmware quirks, specific component layouts — web-search the printer model + the specific component before describing the procedure. General mechanical knowledge from training data is unreliable at the model-specific level: similar printers from the same vendor can have completely different mechanisms (e.g. Kobra 2 has screw-knob tensioners, Mega S has clamp-only fixed belts with no tensioner at all).

When making categorical statements about causation — "X cannot cause Y," "Y is never due to X," "the only cause of Y is Z" — web-search before asserting. Causal claims that exclude a possibility are often wrong because of edge cases or mechanisms the assistant doesn't immediately recall. The cost of saying "let me check" is small; the cost of being confidently wrong erodes trust and sends the user down dead-end diagnostic paths.

Patterns that have caused failure:

  • Asserting that SD card "cannot cause" layer shift (it can, via dropped g-code commands)
  • Describing belt tensioner procedure for a printer that has no tensioner
  • Recommending firmware features that aren't available on the user's specific firmware version

Phrasing for uncertainty:

  • ✅ "I think X is rare, but let me verify before stating it categorically."
  • ✅ "Let me search current sources because I'm not sure if this applies to your specific model."
  • ❌ "X cannot cause Y." (without verification)
  • ❌ "Just adjust the tensioner on the right side." (without verifying that tensioner exists on that model)

This rule applies especially when the user pushes back ("are you sure?") — that pushback is itself a signal to verify, not to dig in defensively.

Examples

Example 1 — minimal input:

  • User: "Just bought ROSA 3D PETG for my Kobra 2 Pro, what settings?"
  • Action: load PETG baseline + Kobra 2 Pro quirks. No problem-specific overrides (general starting profile). Generate .ini. Mention drying as the first thing to do before even printing.

Example 2 — problem-driven:

  • User: "Massive stringing on my Anet ET4 with eSun PLA+, white color"
  • Action: load PLA baseline + Anet quirks if known (or generic Bowden setup if not) + stringing-specific overrides from troubleshooting. Lead the rationale with drying + retraction; profile reflects increased retraction and slightly lower temp.

Example 3 — ambiguous:

  • User: "Make me a profile, my prints look bad"
  • Action: ask 2-3 questions in one batch — material, printer, what specifically looks bad. Don't proceed without these.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.