submission-check
Skill1 starsChanged 2 months agoMITView on GitHub
Gate the three ERC-26 Phase 1 deliverables against every stated requirement before the 2026-09-15 submission. Checks the launch contract, package dependencies, clean-container build, the video's six constraints, and the report's six mandated sections. Use in the final week, and before submitting.
In this repository37 other files it gives its agents.
- wsb-data-pipelineSkill
- docx-exportSkill
- harvard-mdxSkill
- humanise-docx-reportSkill
- make-pdfSkill
- bom-builderSkill
- ring-testSkill
- batch-draftSkill
- cold-emailSkill
- cv-intakeSkill
- cv-upgradeSkill
- find-contactSkill
- followupSkill
- humanize-emailSkill
- log-outreachSkill
- outreach-statusSkill
- target-researchSkill
- erc-statusSkill
- log-sessionSkill
- score-auditSkill
- sumo-statusSkill
- trial-runSkill
- vr-statusSkill
- academic-registerSkill
- aviation-data-sourcesSkill
- business-frameworksSkill
- consultancy-registerSkill
- excel-deliverablesSkill
- forecasting-methodsSkill
- operations-frameworksSkill
- secondary-data-sourcesSkill
- archive-indexSkill
- event-proposalSkill
- meeting-minutesSkill
- read-boxnoteSkill
- captureSkill
- new-scriptSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 49
Every item is individually disqualifying.
- Distinctive: line 115
Check the video's duration and speed with a tool,: not by eye. Both are trivially checkable by the committee and both are disqualifying.
- Distinctive: line 117
If the submission form has not been located yet, flag it.
≈ “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.
Files
/submission-check, deliverables gate
The last thing that runs before submitting. Its job is to catch the defects that make good work score zero.
Two failure modes it exists for, both stated by the organisers as absolute:
- A submission that depends on environment modifications, or does not run correctly inside the provided Docker image, will not be evaluated.
- A submission whose launch file does not accept
shelf_column_numberandbook_colourcorrectly will not be evaluated.
Neither is a deduction. Both discard everything.
Steps
Work through all four blocks. Report every unchecked item, and do not summarise a partial pass as a pass.
Block 1: evaluation risk, check these first
Anything failing here means the submission is not graded at all.
solution.launch.pyaccepts both arguments, tested with the literal organiser command:ros2 launch <package> solution.launch.py shelf_column_number:=2 book_colour:=red- Every dependency declared in
package.xml, verified by a fresh-clone build in a fresh container, not by reading the file - No modification to the provided Docker environment: OS, ROS 2 distribution, Gazebo version, robot model
- Fresh clone into a fresh container:
colcon buildclean with zero manual fixes - Runs on a second machine, if one exists
- Repository is public and the link resolves without a login
- Scored topic names and types exactly as specified, never renamed:
/erc/shelf_column_identificationand/erc/shelf_row_identification, bothstd_msgs/msg/Int32
Block 2: code, 20%
- Perception, planning and execution in distinct nodes, not one file
README.mdin the repository describing the packages the team created- Comments explaining functions and sections
- Every stage has a timeout and a retry budget
- Recovery paths exist for missed grasps, lost topics and unexpected sensor input, and each has been deliberately triggered at least once
- No dead code, no commented-out blocks
- All tunables in
config/params.yaml, none embedded in nodes - No hard-coded shelf pose, row height, or digit-to-column mapping
mainbuilds clean and is the commit being submitted
Block 3: video, 10%
Every item is individually disqualifying.
- Duration 5:00 or under
- Single continuous recording, no cuts
- Not sped up, including subtle speed changes
- On-screen timer visible from trial start to finish
- Team name visible before the trial begins
- Opens by launching
simulation.launch.py, thensolution.launch.py - Launch command shows both arguments
- Every scored task visible in frame
- Uploaded to YouTube, link accessible without login
- The recorded code matches the submitted commit
- Consistent with what the report claims
Block 4: report, 20%
- PDF format
- Rendered page count is 5 or fewer, excluding the title page. Check the render, not the source
- Section 1, system architecture, with a node diagram
- Section 2, perception, with quantitative accuracy across trials
- Section 3, navigation, with travel time data and RViz planned-versus-executed screenshots
- Section 4, manipulation
- Section 5, results, with scores across 5 trials, detection and grasp success rates, average time
- Section 6, limitations and future work
- Six sections in the mandated order, used as literal headings
- Every number traceable to
results/trials.csv - No claim that a test produced a result it did not
- Humanizer pass applied, with all measurements, tables and citations preserved
Clean Technical Documentstandard applied, and every rendered page inspected for blank pages, clipping, overlap and table borders
Block 5: the artifacts on disk
erc_images/contains images from real trial runs, each timestamped, each with a visible bounding box- No stale images from outside a trial run
results/trials.csvhas at least the five report trials, logged with no parameter changes between them- The submission form itself has been located and filled. The organisers provide it closer to the deadline
Output format
## Submission check, <date>, <days> days to 2026-09-15
### Evaluation risk
<PASS, or the failing items. Anything here blocks everything>
### Code (20%)
<n>/<n> checks pass. Failing: <list>
### Video (10%)
<n>/<n> checks pass. Failing: <list>
### Report (20%)
<n>/<n> checks pass. Failing: <list>
### Artifacts
<n>/<n> checks pass. Failing: <list>
## Verdict
<READY TO SUBMIT, or the ordered list of what must be fixed first>Rules
- Report evaluation risk separately and first. A perfect solution that is not evaluated scores zero, so that class of defect outranks every other finding regardless of how small the fix is.
- Verify, don't read. The dependency check means building from a fresh clone in a fresh container. Reading
package.xmlproves nothing about what was installed by hand months ago. - Check the rendered page count. The 5-page limit applies to the PDF, and a source document that looks short can render to six pages.
- Check the video's duration and speed with a tool, not by eye. Both are trivially checkable by the committee and both are disqualifying.
- Never report a partial pass as a pass. List every unchecked item.
- If the submission form has not been located yet, flag it. It is an external dependency and the organisers release it late.
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
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.

