aalayed commited on
Commit
a46ae2c
·
verified ·
1 Parent(s): 688d042

Card: remove references to the unpublished code repository

Browse files
Files changed (2) hide show
  1. CITATION.cff +1 -1
  2. README.md +7 -7
CITATION.cff CHANGED
@@ -18,7 +18,7 @@ keywords:
18
  - input-classification
19
  - indirect-prompt-injection
20
  - multilingual
21
- repository-code: "https://github.com/Abdulaziz1988/llm-prompt-injection-ext"
22
  references:
23
  - type: dataset
24
  title: "CEDA-215: Command-Execution Defense Assessment"
 
18
  - input-classification
19
  - indirect-prompt-injection
20
  - multilingual
21
+ url: "https://huggingface.co/datasets/aalayed/CEDA-1K"
22
  references:
23
  - type: dataset
24
  title: "CEDA-215: Command-Execution Defense Assessment"
README.md CHANGED
@@ -143,16 +143,16 @@ Multi-turn turns are labeled individually: preparatory turns are SAFE on their o
143
 
144
  ## Source and construction
145
 
146
- 1. **Base.** CEDA-215 is copied unchanged from `data/release/CEDA-215/ceda-215.json`.
147
- 2. **Generation.** `scripts/generateCorpus.js` made one Claude API request per taxonomy cell with **`claude-opus-4-8`** (adaptive thinking, effort `high`, structured JSON output), 88 cells in total, 0 refused. The system prompt carried the whitelist, the labeling policy above, the cell definition, the required mix of language, register, length and typos, and the nearest CEDA-215 inputs as examples not to paraphrase. `claude-opus-5` was tried first but its safety filter refused attack-generation cells (category `cyber`); no fallback model was used, so every entry has the same generator.
148
  3. **Sanity checks** (fail the entry, log it): SAFE entries contain none of `;`, `&&`, `||`, `|`, `$(`, backticks, `../`; UNSAFE entries the generator marked `explicitCommand: false` contain no forbidden command token verbatim; the generator label must equal the cell's label; multi-turn control sequences are SAFE on every turn, non-control sequences end UNSAFE, turn counts match. 4 entries were rejected.
149
  4. **Dedupe.** Normalized exact duplicates and pairs with word-unigram Jaccard ≥ 0.8 were dropped, checked against the new pool and all CEDA-215 inputs.
150
- 5. **Trim.** `npm run dataset:build:v2` merges base + corpus and trims each new category to its target (`nl_manipulation` takes the remainder so the total is exactly 1000) by repeatedly dropping the highest-id entry from the largest cell, which keeps cells balanced and the build deterministic.
151
- 6. **Human review.** A seeded, stratified 10 % sample (n = 90 of 888 generated entries) was labeled blind by the author. Agreement with the generator was **91.1 %, Cohen's κ = 0.807**. All 8 disagreements were adjudicated against the written policy; each generator label matched it (per-turn multi-turn labels, listing `/home` or `/tmp`, discussing a forbidden command or sensitive path), so no labels changed. Multi-turn turns were shown to the reviewer on their own, without the earlier turns of their sequence, so 5 of the 8 disagreements are per-turn context cases and κ is conservative. The review files are in `data/corpus/review/`.
152
 
153
  ## Citation
154
 
155
- A persistent identifier for CEDA-1K will be added on publication. Until then, cite the companion repository and the paper:
156
 
157
  ```bibtex
158
  @misc{ceda1k_2026,
@@ -160,7 +160,7 @@ A persistent identifier for CEDA-1K will be added on publication. Until then, ci
160
  author = {Alayed, Abdulaziz},
161
  year = {2026},
162
  version = {1.0.0},
163
- howpublished = {\url{https://github.com/Abdulaziz1988/llm-prompt-injection-ext}}
164
  }
165
  ```
166
 
@@ -179,4 +179,4 @@ CEDA-215 (contained in this dataset): DOI [10.5281/zenodo.20122529](https://doi.
179
 
180
  This dataset is released under the [Creative Commons Attribution 4.0 International License (CC-BY 4.0)](https://creativecommons.org/licenses/by/4.0/). You are free to share and adapt the material for any purpose, including commercial use, provided you give appropriate credit.
181
 
182
- The companion code repository is licensed separately under GPL-3.0-or-later.
 
143
 
144
  ## Source and construction
145
 
146
+ 1. **Base.** CEDA-215 is copied unchanged from its published release.
147
+ 2. **Generation.** A generation script made one Claude API request per taxonomy cell with **`claude-opus-4-8`** (adaptive thinking, effort `high`, structured JSON output), 88 cells in total, 0 refused. The system prompt carried the whitelist, the labeling policy above, the cell definition, the required mix of language, register, length and typos, and the nearest CEDA-215 inputs as examples not to paraphrase. `claude-opus-5` was tried first but its safety filter refused attack-generation cells (category `cyber`); no fallback model was used, so every entry has the same generator.
148
  3. **Sanity checks** (fail the entry, log it): SAFE entries contain none of `;`, `&&`, `||`, `|`, `$(`, backticks, `../`; UNSAFE entries the generator marked `explicitCommand: false` contain no forbidden command token verbatim; the generator label must equal the cell's label; multi-turn control sequences are SAFE on every turn, non-control sequences end UNSAFE, turn counts match. 4 entries were rejected.
149
  4. **Dedupe.** Normalized exact duplicates and pairs with word-unigram Jaccard ≥ 0.8 were dropped, checked against the new pool and all CEDA-215 inputs.
150
+ 5. **Trim.** A deterministic build step merges base + corpus and trims each new category to its target (`nl_manipulation` takes the remainder so the total is exactly 1000) by repeatedly dropping the highest-id entry from the largest cell, which keeps cells balanced and the build deterministic.
151
+ 6. **Human review.** A seeded, stratified 10 % sample (n = 90 of 888 generated entries) was labeled blind by the author. Agreement with the generator was **91.1 %, Cohen's κ = 0.807**. All 8 disagreements were adjudicated against the written policy; each generator label matched it (per-turn multi-turn labels, listing `/home` or `/tmp`, discussing a forbidden command or sensitive path), so no labels changed. Multi-turn turns were shown to the reviewer on their own, without the earlier turns of their sequence, so 5 of the 8 disagreements are per-turn context cases and κ is conservative.
152
 
153
  ## Citation
154
 
155
+ A persistent identifier for CEDA-1K will be added on publication. Until then, cite this dataset page and the paper:
156
 
157
  ```bibtex
158
  @misc{ceda1k_2026,
 
160
  author = {Alayed, Abdulaziz},
161
  year = {2026},
162
  version = {1.0.0},
163
+ howpublished = {\url{https://huggingface.co/datasets/aalayed/CEDA-1K}}
164
  }
165
  ```
166
 
 
179
 
180
  This dataset is released under the [Creative Commons Attribution 4.0 International License (CC-BY 4.0)](https://creativecommons.org/licenses/by/4.0/). You are free to share and adapt the material for any purpose, including commercial use, provided you give appropriate credit.
181
 
182
+ The code used to build and evaluate the dataset is not part of this release and will be published separately.