# Sdrf:fix

> Use when the user has an SDRF file with known errors and wants them fixed automatically. Triggers on requests to fix, correct, or repair SDRF errors.

- **Type:** Skill
- **Install:** `agentstack add skill-bigbio-sdrf-skills-sdrf-fix`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [bigbio](https://agentstack.voostack.com/s/bigbio)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [bigbio](https://github.com/bigbio)
- **Source:** https://github.com/bigbio/sdrf-skills/tree/main/skills/sdrf-fix
- **Website:** https://sdrf.quantms.org

## Install

```sh
agentstack add skill-bigbio-sdrf-skills-sdrf-fix
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# SDRF Auto-Fix Workflow

You are fixing known common errors in an SDRF file. Apply fixes systematically.

## Step 0: Check parse_sdrf availability

Verify that `parse_sdrf` is available (run `parse_sdrf --version` or `which parse_sdrf`). If it is not installed:
- Inform the user that re-validation after fixes will need to be done manually
- Suggest `/sdrf:setup` or `conda env create -f environment.yml && conda activate sdrf-skills` (or `pip install -r requirements.txt`)
- Continue with the fixes; the user can validate later once dependencies are installed

## Common Error Patterns and Their Fixes

### 1. UNIMOD Accession Swaps (45% of all errors)
| Wrong | Correct | Modification |
|-------|---------|-------------|
| UNIMOD:21 for Acetyl | UNIMOD:1 | Acetyl |
| UNIMOD:1 for Phospho | UNIMOD:21 | Phospho |
| UNIMOD:34 for Oxidation | UNIMOD:35 | Oxidation |
| UNIMOD:35 for Methyl | UNIMOD:34 | Methyl |

**Fix**: Parse NT= field, look up correct UNIMOD accession, replace AC= field.

### 2. Missing Ontology Prefix (30%)
| Wrong | Correct |
|-------|---------|
| `0000305` | `EFO:0000305` |
| `9606` | `NCBITaxon:9606` |
| `0002107` | `UBERON:0002107` |

**Fix**: Detect bare numbers, infer ontology from column type, add prefix.

### 3. Case Normalization (25%)
| Wrong | Correct |
|-------|---------|
| `Male` | `male` |
| `Female` | `female` |
| `Homo Sapiens` | `Homo sapiens` |
| `Not Available` | `not available` |

**Fix**: Lowercase sex values and reserved words. Organism names follow binomial rules (capital genus, lowercase species).

### 4. Python/Programming Artifacts (15%)
| Wrong | Correct |
|-------|---------|
| `['breast cancer']` | `breast cancer` |
| `nan` | `not available` |
| `None` | `not available` |
| `""` | (empty or `not available`) |

**Fix**: Strip brackets, quotes; replace nan/None with reserved words.

### 5. Reserved Word Standardization
| Wrong | Correct |
|-------|---------|
| `N/A` | `not applicable` |
| `NA` | `not available` |
| `n/a` | `not applicable` |
| `unknown` | `not available` |
| `null` | `not available` |
| `-` | `not available` |

**Fix**: Replace with correct SDRF reserved words.

### 6. DIA/DDA Terminology
| Wrong | Correct |
|-------|---------|
| `data-dependent acquisition` | `Data-Dependent Acquisition` |
| `data-independent` | `Data-Independent Acquisition` |
| `DDA` | `Data-Dependent Acquisition` |
| `DIA` | `Data-Independent Acquisition` |

**Fix**: Use the ontology-standard full name.

### 7. Age Format
| Wrong | Correct |
|-------|---------|
| `58 years` | `58Y` |
| `58` | `58Y` |
| `6 months` | `6M` |
| `14 days` | `14D` |
| `58yo` | `58Y` |

**Fix**: Extract number, map unit to Y/M/D suffix.

### 8. Modification Parameter Format
| Wrong | Correct |
|-------|---------|
| `Carbamidomethyl (C)` | `NT=Carbamidomethyl;AC=UNIMOD:4;TA=C;MT=Fixed` |
| `Oxidation (M)` | `NT=Oxidation;AC=UNIMOD:35;TA=M;MT=Variable` |
| `NT=Acetyl;AC=UNIMOD:1;TA=Protein N-term` | `NT=Acetyl;AC=UNIMOD:1;PP=Protein N-term;MT=Variable` |

**Fix**: Parse free-text mods, construct proper NT/AC/TA/MT format.

### 9. Trailing Whitespace
**Fix**: Trim all cell values and column names.

### 10. Instrument Format
| Wrong | Correct |
|-------|---------|
| `Q Exactive` | `AC=MS:1001911;NT=Q Exactive` |
| `Orbitrap Fusion Lumos` | `AC=MS:1002732;NT=Orbitrap Fusion Lumos` |

**Fix**: If missing AC= format, search OLS MS ontology and construct proper format.
```text
mcp OLS → searchClasses(query="", ontologyId="ms")
```

## Fix Procedure

1. **Parse** the SDRF into a structured table
2. **Scan** every cell for each error pattern above
3. **Apply fixes** — for ontology-dependent fixes, verify via OLS before changing
4. **Log changes** — track every change made (row, column, old value, new value, reason)
5. **Present changelog** to user before outputting the fixed SDRF
6. **Output** the corrected SDRF as a TSV code block

## Changelog Format

```text
Changes Applied:
  Row 3, comment[modification parameters]:
    OLD: NT=Acetyl;AC=UNIMOD:21;TA=Protein N-term;MT=Variable
    NEW: NT=Acetyl;AC=UNIMOD:1;PP=Protein N-term;MT=Variable
    FIX: UNIMOD:21 is Phospho, not Acetyl. Correct accession is UNIMOD:1. Also TA→PP for position.

  Row 5, characteristics[sex]:
    OLD: Male
    NEW: male
    FIX: Sex values must be lowercase per SDRF specification.

  All rows, comment[data file]:
    FIX: Trimmed trailing whitespace from 12 values.

Summary: 15 fixes applied (3 UNIMOD corrections, 5 case fixes, 7 whitespace trims)
```

## Step After Fixes: Re-Validate with sdrf-pipelines

After applying all fixes, **always** run programmatic validation before presenting
results to the user.

### 1. Update spec to latest version
```bash
git submodule update --remote --recursive
```

### 2. Run sdrf-pipelines validation
Save the fixed SDRF to a file and validate with the detected templates:
```bash
parse_sdrf validate-sdrf \
  --sdrf_file fixed.sdrf.tsv \
  --template  \
  --template 
```
Detect templates from `comment[sdrf template]` columns in the SDRF.
If `parse_sdrf` is not installed, tell the user: `pip install sdrf-pipelines`

### 3. Interpret results
1. If validation passes → present the changelog + fixed SDRF to the user
2. If validation finds new errors → fix them and re-run until clean
3. Verify fixed UNIMOD accessions match the NT= modification names
4. Read `spec/sdrf-proteomics/TERMS.tsv` and check `allow_not_available`/`allow_not_applicable` fields
5. Count: total fixes applied, remaining issues not auto-fixable

Present the re-validation summary alongside the changelog.

If all errors are fixed and the SDRF is for a ProteomeXchange dataset (PXD accession),
suggest contributing the corrected annotation via `/sdrf:contribute {PXD}` to the
`sdrf-annotated-datasets` community repository.

## When NOT to Auto-Fix

- Values that might be intentionally different (ask the user)
- Ontology terms where the "correct" version is ambiguous
- Missing columns (suggest but don't add without user approval)
- Factor values (design decisions — always ask)
- Cell line names (need Cellosaurus verification)
- Organism names that might be intentional (e.g., hybrid organisms)

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [bigbio](https://github.com/bigbio)
- **Source:** [bigbio/sdrf-skills](https://github.com/bigbio/sdrf-skills)
- **License:** MIT
- **Homepage:** https://sdrf.quantms.org

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-bigbio-sdrf-skills-sdrf-fix
- Seller: https://agentstack.voostack.com/s/bigbio
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.