{"slug":"hic-aggregation","title":"hic-aggregation","summary":"Build comprehensive chromatin contact maps by aggregating Hi-C loop calls (BEDPE) across multiple ENCODE experiments, donors, and labs. Use when the user wants to answer \"what regions are in 3D contact in my tissue?\" by creating a union catalog of chromatin loops. Handles resolut","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-09-22T13:25:34.440435Z","repo":{"url":"https://github.com/ammawla/encode-toolkit","stars":21,"forks":5,"license":"AGPL-3.0","updatedAt":"2026-09-27T01:22:42Z"},"bodyHtml":"
\n

name: hic-aggregation\ndescription: Build comprehensive chromatin contact maps by aggregating Hi-C loop calls (BEDPE) across multiple ENCODE experiments, donors, and labs. Use when the user wants to answer \"what regions are in 3D contact in my tissue?\" by creating a union catalog of chromatin loops. Handles resolution-aware anchor matching, cross-lab variation, and Hi-C-specific quality metrics.

\n

Aggregate Hi-C Chromatin Contacts Across Studies

\n

When to Use

\n\n

Build a comprehensive catalog of chromatin loops for a tissue/cell type by merging BEDPE loop calls from multiple ENCODE Hi-C experiments.

\n

Scientific Rationale

\n

The question: \"What regions are in 3D physical contact in my tissue?\"

\n

Like histone marks and accessibility, chromatin loops are a detection question. If a loop between Region A and Region B is detected in one donor but not another, the contact is still real — individual variation, sequencing depth, and computational resolution explain absence. We want the union of all detected contacts.

\n

Key Concepts

\n

Hi-C data measures pairwise chromatin interactions genome-wide. After processing:

\n\n

BEDPE format (Paired-End BED):

\n
chr1  start1  end1  chr2  start2  end2  name  score  strand1  strand2\n
\n

Each row represents a contact between two genomic anchor regions.

\n

Literature Support

\n\n

Step 1: Find All Available Hi-C Data

\n
encode_search_experiments(\n    assay_title=\"Hi-C\",\n    organ=\"pancreas\",           # user's tissue of interest\n    biosample_type=\"tissue\",\n    limit=100\n)\n
\n

Present a summary to the user:

\n\n

Use encode_get_facets to check availability:

\n
encode_get_facets(assay_title=\"Hi-C\", organ=\"pancreas\")\n
\n

Note: Hi-C data is computationally expensive to produce, so there are typically fewer experiments per tissue than ChIP-seq or ATAC-seq. Even 2-3 experiments can be valuable for union catalogs.

\n

Step 2: Quality-Gate Each Experiment

\n
encode_get_experiment(accession=\"ENCSR...\")\n
\n

Hi-C Quality Checks

\n\n

Include if:

\n\n

Exclude if:

\n\n

Track all included experiments:

\n
encode_track_experiment(accession=\"ENCSR...\")\n
\n

Step 3: Download Loop Call Files

\n

For each experiment, get BEDPE loop calls:

\n
# Search for loop/interaction files\nencode_list_files(\n    experiment_accession=\"ENCSR...\",\n    file_format=\"bedpe\",\n    assembly=\"GRCh38\"\n)\n\n# Or ask for loop calls by output type\nencode_list_files(\n    experiment_accession=\"ENCSR...\",\n    output_type=\"loops\",\n    assembly=\"GRCh38\"\n)\n\n# Or contact domains\nencode_list_files(\n    experiment_accession=\"ENCSR...\",\n    output_type=\"contact domains\",\n    assembly=\"GRCh38\"\n)\n
\n

File selection priority:

\n
    \n
  1. Chromatin interactions (loop calls from HICCUPS or similar)
    2. \n
  2. Contact domains (TADs — different analysis, handle separately)
    3. \n
  3. Replicated loops (if available)
    4. \n
\n

Prefer preferred_default=True files when available.

\n
encode_download_files(\n    file_accessions=[\"ENCFF...\", ...],\n    download_dir=\"/path/to/data/hic_loops\",\n    organize_by=\"flat\"\n)\n
\n

Validate the downloaded BEDPE files before filtering. The report gives the detected\nanchor resolution, which Step 4 needs; gzipped inputs are read directly.

\n
python3 scripts/validate_loops.py sample.bedpe [--min-distance 20000] [--expected-resolution 10000]\n
\n

Step 4: Understanding Hi-C Resolution and Anchors

\n

Critical: Resolution-Aware Processing

\n

Hi-C loop anchors are binned regions, not precise positions. The resolution determines anchor size:

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
ResolutionAnchor WidthBest ForTypical Loop Count
5 kb5,000 bpFine-scale promoter-enhancer loopsMore loops
10 kb10,000 bpStandard analysisModerate
25 kb25,000 bpLarge-scale domain contactsFewer loops
\n

All loops being merged must be at the same resolution, or anchors must be harmonized to a common resolution.

\n

Harmonizing Resolution

\n

If experiments have loops called at different resolutions:

\n
# Expand 5kb anchors to 10kb resolution\nawk -v res=10000 'BEGIN{OFS=\"\\t\"} {\n    # Bin anchor 1\n    bin1_start = int($2/res) * res\n    bin1_end = bin1_start + res\n    # Bin anchor 2\n    bin2_start = int($5/res) * res\n    bin2_end = bin2_start + res\n    print $1, bin1_start, bin1_end, $4, bin2_start, bin2_end, $7, $8, $9, $10\n}' fine_res_loops.bedpe > harmonized_loops.bedpe\n
\n

Step 5: Per-Sample Filtering

\n

5a. ENCODE Blocklist Filtering (Amemiya et al. 2019)

\n

Remove loops with anchors in artifact-prone regions (download from https://github.com/Boyle-Lab/Blacklist/blob/master/lists/hg38-blacklist.v2.bed.gz):

\n
# Filter loops where EITHER anchor overlaps a blocklist region\ngunzip -k hg38-blacklist.v2.bed.gz\n\n# First, extract anchor 1 and anchor 2 as separate BED files\nawk 'BEGIN{OFS=\"\\t\"} {print $1,$2,$3,NR}' sample.bedpe > anchors1.bed\nawk 'BEGIN{OFS=\"\\t\"} {print $4,$5,$6,NR}' sample.bedpe > anchors2.bed\n\n# Find anchor rows NOT in blocklist\nbedtools intersect -a anchors1.bed -b hg38-blacklist.v2.bed -v | cut -f4 > clean_rows_1.txt\nbedtools intersect -a anchors2.bed -b hg38-blacklist.v2.bed -v | cut -f4 > clean_rows_2.txt\n\n# Keep only rows where BOTH anchors pass\ncomm -12 <(sort clean_rows_1.txt) <(sort clean_rows_2.txt) > clean_rows.txt\nawk 'NR==FNR{a[$1];next} FNR in a' clean_rows.txt sample.bedpe > sample.filtered.bedpe\n
\n

5b. Score Filtering

\n

Filter by interaction score/significance:

\n
# If BEDPE has a score column (col 8), filter to significant interactions\n# Keep top 75% by score (true distribution quantile, not range-based)\nTOTAL=$(wc -l < sample.filtered.bedpe)\nLINE_25=$(echo \"$TOTAL\" | awk '{printf \"%d\", $1 * 0.25}')\nTHRESHOLD=$(sort -k8,8n sample.filtered.bedpe | awk -v line=\"$LINE_25\" 'NR==line{print $8}')\nawk -v t=\"$THRESHOLD\" '$8 >= t' sample.filtered.bedpe > sample.qfiltered.bedpe\n
\n

5c. Remove Self-Ligation Artifacts

\n

Loops where both anchors are very close are likely artifacts:

\n
# Remove loops where anchors are on same chromosome and < 20kb apart\nawk '{\n    if ($1 != $4) print $0;  # inter-chromosomal: keep (rare but real)\n    else if (($5 - $3) >= 20000) print $0;  # > 20kb apart: keep\n}' sample.qfiltered.bedpe > sample.clean.bedpe\n
\n

Step 6: Union Merge of Loops

\n

The Paired-Region Matching Problem

\n

Unlike peaks (single regions), loops are pairs of regions. Two loops match if both anchors overlap:

\n
Loop 1:  [anchor1A]--------[anchor1B]\nLoop 2:    [anchor2A]------[anchor2B]\n
\n

These should merge if anchor1A overlaps anchor2A AND anchor1B overlaps anchor2B.

\n

Method A: bedtools pairToPair (Recommended for simple union)

\n
# Concatenate all filtered loops\ncat sample1.clean.bedpe sample2.clean.bedpe ... > all_loops.bedpe\n\n# Sort by anchor 1 coordinates\nsort -k1,1 -k2,2n -k4,4 -k5,5n all_loops.bedpe > all_loops.sorted.bedpe\n\n# Use a custom merge approach:\n# 1. Bin anchors to resolution, creating a loop ID\n# 2. Group by loop ID\n# 3. Count support\n\nawk -v res=10000 'BEGIN{OFS=\"\\t\"} {\n    # Create binned anchor coordinates as loop identifier\n    a1_bin = $1 \":\" int($2/res)*res\n    a2_bin = $4 \":\" int($5/res)*res\n    # Canonical order (smaller coordinate first) to handle orientation\n    if (a1_bin < a2_bin) loop_id = a1_bin \"-\" a2_bin\n    else loop_id = a2_bin \"-\" a1_bin\n    print loop_id, $0\n}' all_loops.sorted.bedpe | \\\nsort -k1,1 | \\\nawk 'BEGIN{OFS=\"\\t\"} {\n    if ($1 != prev_id) {\n        if (NR > 1) print chr1, start1, end1, chr2, start2, end2, count, max_score\n        prev_id = $1\n        chr1=$2; start1=$3; end1=$4; chr2=$5; start2=$6; end2=$7\n        count = 1; max_score = $9\n    } else {\n        count++\n        if ($9 > max_score) max_score = $9\n        # Expand anchors to encompass all overlapping calls\n        if ($3 < start1) start1 = $3\n        if ($4 > end1) end1 = $4\n        if ($6 < start2) start2 = $6\n        if ($7 > end2) end2 = $7\n    }\n} END {\n    print chr1, start1, end1, chr2, start2, end2, count, max_score\n}' > union_loops.bedpe\n
\n

Method B: Resolution-Binned Approach (Loop Catalog method)

\n

Following the Loop Catalog (Reyna et al. 2025) approach:

\n
# Bin all loop anchors to a fixed resolution\nawk -v res=10000 'BEGIN{OFS=\"\\t\"} {\n    a1_start = int($2/res) * res\n    a1_end = a1_start + res\n    a2_start = int($5/res) * res\n    a2_end = a2_start + res\n    # Canonical ordering\n    if ($1 < $4 || ($1 == $4 && a1_start <= a2_start))\n        print $1, a1_start, a1_end, $4, a2_start, a2_end\n    else\n        print $4, a2_start, a2_end, $1, a1_start, a1_end\n}' all_loops.sorted.bedpe | \\\nsort -u | \\\nsort -k1,1 -k2,2n -k4,4 -k5,5n | \\\nuniq -c | \\\nawk 'BEGIN{OFS=\"\\t\"} {print $2,$3,$4,$5,$6,$7,$1}' > union_loops_binned.bedpe\n# Columns: chr1, start1, end1, chr2, start2, end2, n_supporting_samples\n
\n

Method C: Using Specialized Tools

\n

mariner (R/Bioconductor):

\n
library(mariner)\n# Read BEDPE files as GInteractions\nloops <- lapply(bedpe_files, read.table)\n# Convert to GInteractions and merge\ngi <- as_ginteractions(loops)\nmerged <- mergePairs(gi, radius = 10000)  # 10kb tolerance\n
\n

AQuA Tools (Python):

\n
# BEDPE union with anchor overlap tolerance\naqua bedpe-union -i sample1.bedpe sample2.bedpe -o union.bedpe --slop 5000\n
\n

Step 7: Confidence Annotation

\n

Given N total experiments:

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
ConfidenceCriteriaInterpretation
HighDetected in >=50% of samplesConstitutive loop, present across individuals
SupportedDetected in 2+ samplesLikely real, some variation
SingletonDetected in 1 sample onlyMay be individual-specific or depth-dependent
\n
awk -v N=4 '{\n    if ($7 >= N*0.5) conf=\"HIGH\";\n    else if ($7 >= 2) conf=\"SUPPORTED\";\n    else conf=\"SINGLETON\";\n    print $0\"\\t\"conf\"\\t\"$7\"/\"N\n}' union_loops_binned.bedpe > union_loops.annotated.bedpe\n
\n

Context for singletons: Hi-C loop detection is very sensitive to sequencing depth. Many singletons may simply be under-powered in other samples rather than biologically absent. The Loop Catalog found that a union approach captures ~3x more loops than any individual experiment.

\n

Step 8: Separate Analysis for TADs and Compartments

\n

TAD boundaries and A/B compartments require different aggregation than loops:

\n

TAD Boundaries

\n

TAD boundaries are single genomic positions. Aggregate like narrow peaks:

\n
# Extract TAD boundary BED from contact domain files\n# Each boundary is a narrow region\ncat tad_boundaries_sample*.bed | \\\nbedtools sort -i - | \\\nbedtools merge -i - -d 40000 -c 1 -o count > union_tad_boundaries.bed\n# 40kb gap tolerance because TAD boundaries are resolution-dependent\n
\n

A/B Compartments

\n

Compartment calls (eigenvector sign at each bin) should be aggregated by majority vote:

\n
# For each resolution bin, assign A or B based on majority of samples\n# This is more complex and typically done in R/Python\n
\n

Step 9: Log Provenance

\n
encode_log_derived_file(\n    file_path=\"/path/to/union_loops.annotated.bedpe\",\n    source_accessions=[\"ENCSR...\", \"ENCSR...\", ...],\n    description=\"Union chromatin loops across N pancreas Hi-C experiments\",\n    file_type=\"aggregated_loops\",\n    tool_used=\"bedtools + custom merge at 10kb resolution\",\n    parameters=\"blocklist filtered, score >= 25th pctl, self-ligation >= 20kb removed, 10kb resolution binning\"\n)\n
\n

Step 10: Summary Statistics

\n

Report to the user:

\n\n

Pitfalls Specific to Hi-C Data

\n
    \n

  1. Resolution mismatch: Loop calls at 5kb vs 25kb resolution will have very different anchor sizes. Always harmonize to a common resolution before merging.

    \n
    2. \n

  2. Sequencing depth sensitivity: Loop calling requires deep sequencing (400M+ valid pairs). Shallowly sequenced experiments will call far fewer loops — this is under-detection, not absence.

    \n
    3. \n

  3. Algorithm differences are LARGE: Wolff et al. 2022 (GigaScience) found that HICCUPS, Mustache, Fit-Hi-C, and HiCExplorer loop callers intersect by ~50% at most. Mustache tends to recover more validated loops (Roayaei Ardakany et al. 2020). If mixing callers, note this in provenance — and this discordance is itself a reason to prefer the union approach.

    \n
    4. \n

  4. Orientation matters: BEDPE anchors should be canonically ordered (anchor1 < anchor2 by genomic coordinate) before merging to avoid duplicate counting.

    \n
    5. \n

  5. Inter-chromosomal contacts: These are rare but real. Handle separately — they cannot be distance-filtered.

    \n
    6. \n

  6. Distance distribution: Most loops are 100kb-2Mb. Very short-range contacts (<20kb) are often noise from undigested chromatin. Very long-range (>10Mb) are rare.

    \n
    7. \n

  7. Do NOT mix assemblies: All files must be GRCh38 or all hg19. Hi-C resolution binning makes liftOver of loops particularly error-prone.

    \n
    8. \n

  8. TADs vs loops: These are different features. TADs are domains (regions), loops are point contacts (pairs). Do not mix them in the same union.

    \n
    9. \n

  9. Micro-C as complement: Micro-C achieves higher resolution than Hi-C and can detect sub-TAD loops. Treat Micro-C loops as compatible with Hi-C loops in a union (Mustache works on both).

    \n
    10. \n
\n

Walkthrough: Building a Cross-Tissue Loop Catalog for the MYC Locus

\n

Goal: Aggregate Hi-C chromatin loops across tissues to identify conserved and tissue-specific 3D contacts at the MYC gene locus.\nContext: Cancer research — MYC is regulated by distal enhancers via chromatin looping.

\n

Step 1: Find Hi-C experiments across tissues

\n
encode_search_experiments(assay_title=\"Hi-C\", organism=\"Homo sapiens\", limit=50)\n
\n

Expected output:

\n
{\n  \"results\": [\n    {\"accession\": \"ENCSR000AKA\", \"assay_title\": \"Hi-C\", \"biosample_summary\": \"GM12878\", \"status\": \"released\"},\n    {\"accession\": \"ENCSR489OCU\", \"assay_title\": \"Hi-C\", \"biosample_summary\": \"K562\", \"status\": \"released\"},\n    {\"accession\": \"ENCSR382RFU\", \"assay_title\": \"Hi-C\", \"biosample_summary\": \"liver\", \"status\": \"released\"}\n  ],\n  \"total\": 89,\n  \"limit\": 50,\n  \"offset\": 0,\n  \"has_more\": true,\n  \"next_offset\": 50\n}\n
\n

Interpretation: 89 Hi-C experiments available. Select 5–10 spanning diverse tissue types for cross-tissue comparison.

\n

Step 2: List loop files for each experiment

\n
encode_list_files(experiment_accession=\"ENCSR000AKA\", file_format=\"bedpe\", assembly=\"GRCh38\")\n
\n

Expected output (a JSON array of file records; fields abridged):

\n
[\n  {\"accession\": \"ENCFF001ABC\", \"output_type\": \"contact domains\", \"file_format\": \"bedpe\", \"file_size_human\": \"2.4 MB\"},\n  {\"accession\": \"ENCFF002DEF\", \"output_type\": \"loops\", \"file_format\": \"bedpe\", \"file_size_human\": \"1.8 MB\"}\n]\n
\n

Interpretation: Use \"loops\" files for loop aggregation. Contact domains are TADs, not loops.

\n

Step 3: Download loop files

\n
encode_download_files(file_accessions=[\"ENCFF002DEF\", \"ENCFF003GHI\", \"ENCFF004JKL\"], download_dir=\"/data/hic_loops\")\n
\n

Expected output (one of the three downloaded entries shown):

\n
{\n  \"downloaded\": [\n    {\n      \"accession\": \"ENCFF002DEF\",\n      \"file_path\": \"/data/hic_loops/ENCFF002DEF.bedpe\",\n      \"file_size\": 1887436,\n      \"file_size_human\": \"1.8 MB\",\n      \"success\": true,\n      \"error\": \"\",\n      \"md5_verified\": true\n    }\n  ],\n  \"errors\": [],\n  \"summary\": {\n    \"total_requested\": 3,\n    \"successful\": 3,\n    \"failed\": 0,\n    \"total_size\": 5872025,\n    \"total_size_human\": \"5.6 MB\"\n  }\n}\n
\n

Step 4: Aggregate loops with resolution-aware anchor matching

\n

Apply union merge across tissues:

\n\n

Step 5: Filter to MYC locus

\n
# MYC locus: chr8:127,700,000-128,000,000\nawk '$1==\"chr8\" && $2>=127700000 && $3<=128000000' union_loops.bedpe > myc_loops.bedpe\n
\n

Interpretation: Loops anchored at the MYC promoter connecting to distal enhancers. Conserved loops (≥3 tissues) likely represent fundamental regulatory architecture; tissue-specific loops may drive context-dependent MYC activation.

\n

Integration with downstream skills

\n\n

Code Examples

\n

1. Survey available Hi-C data by tissue

\n
encode_get_facets(assay_title=\"Hi-C\", organism=\"Homo sapiens\")\n
\n

Expected output (facet field names are the top-level keys):

\n
{\n  \"biosample_ontology.organ_slims\": [\n    {\"term\": \"brain\", \"count\": 24},\n    {\"term\": \"blood\", \"count\": 15}\n  ]\n}\n
\n

2. Get details for a specific Hi-C experiment

\n
encode_get_experiment(accession=\"ENCSR000AKA\")\n
\n

Expected output (fields abridged):

\n
{\n  \"accession\": \"ENCSR000AKA\",\n  \"assay_title\": \"Hi-C\",\n  \"biosample_summary\": \"GM12878\",\n  \"assembly\": [\"GRCh38\"],\n  \"bio_replicate_count\": 2,\n  \"status\": \"released\",\n  \"lab\": \"Erez Lieberman Aiden, Baylor\",\n  \"audit_error_count\": 0,\n  \"audit_not_compliant_count\": 0,\n  \"audit_warning_count\": 1,\n  \"audit_internal_action_count\": 0\n}\n
\n

3. Compare loop sets between two cell types

\n

Both experiments must already be tracked; otherwise the tool returns {\"error\": \"Experiment ... not tracked. Track it first.\"}.

\n
encode_compare_experiments(accession1=\"ENCSR000AKA\", accession2=\"ENCSR489OCU\")\n
\n

Expected output:

\n
{\n  \"experiment_1\": {\"accession\": \"ENCSR000AKA\", \"assay\": \"Hi-C\", \"biosample\": \"GM12878\"},\n  \"experiment_2\": {\"accession\": \"ENCSR489OCU\", \"assay\": \"Hi-C\", \"biosample\": \"K562\"},\n  \"verdict\": \"COMPATIBLE_WITH_CAVEATS\",\n  \"recommendation\": \"These experiments can be compared, but the warnings should be addressed in your analysis.\",\n  \"compatible_aspects\": [\n    \"Same organism: Homo sapiens\",\n    \"Same assembly: GRCh38\",\n    \"Same assay: Hi-C\",\n    \"Same biosample type: cell line\"\n  ],\n  \"issues\": [],\n  \"warnings\": [\n    \"Different labs: Erez Lieberman Aiden, Baylor vs Job Dekker, UMass. Batch effects possible.\"\n  ]\n}\n
\n

Integration

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
This skill produces...Feed into...Purpose
Union loop catalog (BEDPE)peak-annotationAssign genes to loop anchors
Conserved loop coordinateshistone-aggregationOverlay H3K27ac at anchors to find active enhancer-promoter loops
Tissue-specific loopsaccessibility-aggregationCheck if loop anchors overlap open chromatin
Loop anchor BED intervalsvariant-annotationFind GWAS/clinical variants disrupting loop anchors
Loop anchor coordinatesliftover-coordinatesConvert hg19 loops to GRCh38
Aggregated loop statisticsvisualization-workflowGenerate loop frequency heatmaps
Loop-gene assignmentsdisease-researchConnect loop disruptions to disease phenotypes
\n

Related Skills

\n\n

Presenting Results

\n\n

For the request: \"$ARGUMENTS\"

\n","files":[{"path":"references/literature.md","sizeBytes":7356,"isText":true},{"path":"references/loop-caller-comparison.md","sizeBytes":6845,"isText":true},{"path":"scripts/validate_loops.py","sizeBytes":11930,"isText":true},{"path":"SKILL.md","sizeBytes":23379,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-22T13:25:59.188332Z","sha256":"50DA749F8776646C91D7F4DDA8E43B569595969D1EE26906031AF8B23532F428","sizeBytes":18981},"review":null,"source":{"repositoryUrl":"https://github.com/ammawla/encode-toolkit","path":"plugin/skills/hic-aggregation","license":"AGPL-3.0","commit":"36836c8725fd4d20d9c851ce314f5151aea5f57c","subtreeSha":"DC2F069407FEE8F561C9C32CAF37086E685FBE373725D6C8F026746C5167C70A","lastSyncedAt":"2026-09-29T20:56:54.045383Z"},"reviewedAt":"2026-09-22T13:26:50.233209Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/ammawla/encode-toolkit/tree/main/plugin/skills/hic-aggregation"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install ammawla-encode-toolkit@llmmart"},{"target":"git","command":"git clone https://github.com/ammawla/encode-toolkit.git"}]}