Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Build Code Knowledge Graph · athola | DecimalAI

athola/graph-build

NewSkillScoreMaintained

Builds or updates the code knowledge graph via tree-sitter AST and SQLite. Use when setting up the graph before search or blast-radius analysis.

codingby athola

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/athola-graph-build/SKILL.md

decimalai skills pull athola-graph-build --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/athola-graph-build/SKILL.md https://app.decimal.ai/s/athola-graph-build/SKILL.md

Raw file: /s/athola-graph-build/SKILL.md ↗ · machine-readable summary at /s/athola-graph-build.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 3h ago

—-17% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-12	✗→✓	▲ Improved	-48%	0%
case-04	✗→✓	▲ Improved	538%	0%
case-05	✗→✓	▲ Improved	-11%	0%
case-07	✗→✓	▲ Improved	21%	0%
case-08	✗→✓	▲ Improved	17%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

athola / claude-night-market

plugins/gauntlet/skills/graph-build/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1

Skill contents

SKILL.md

Build Code Knowledge Graph

Build or update the .gauntlet/graph.db knowledge graph for the current codebase.

Steps

Detect target: Use the current working directory

or a user-specified path.

Check for existing graph: If .gauntlet/graph.db

exists, run an incremental update. Otherwise, run a full build.

Run the build script:

For full build: bash python3 ${CLAUDE_PLUGIN_ROOT}/scripts/graph_build.py <dir>

For incremental update: bash python3 ${CLAUDE_PLUGIN_ROOT}/scripts/graph_build.py <dir> --incremental

Report results: Show the JSON output including

files parsed, nodes created, edges created, and duration.

Suggest next steps: Recommend searching the graph

or running blast radius analysis.

When To Use

At the start of a session to build structural awareness
After significant code changes to update the graph
Before running blast radius analysis or flow tracing
When the user asks about codebase structure

When NOT To Use

Querying a graph that already exists (use gauntlet:graph-search)
Building the challenge knowledge base (use gauntlet:extract)

What Gets Parsed

The graph extracts nodes (File, Class, Function, Type, Test) and edges (CALLS, IMPORTS_FROM, INHERITS, CONTAINS, IMPLEMENTS, TESTED_BY) from 20+ languages including Python, JavaScript, TypeScript, Go, Rust, Java, C/C++, Ruby, and PHP.

Storage

Database: .gauntlet/graph.db (SQLite with WAL mode)
Auto-creates .gauntlet/.gitignore to prevent commits
Incremental updates use SHA-256 hashing to skip

unchanged files

Exit Criteria

] .gauntlet/graph.db exists and is a valid SQLite file after

the skill completes; .gauntlet/.gitignore exists to prevent the database from being committed

] Build report JSON includes all four keys: files_parsed,

nodes_created, edges_created, duration; values are non-zero for non-empty codebases

] Incremental update path used when .gauntlet/graph.db already

exists (script invoked with --incremental); full build only on first run

] Next-step suggestions presented to the user: search the graph

or run blast-radius analysis

Want to see if athola/graph-build would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/20/2026, 22:40 UTC · 22 cases

Measured on 18 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+50%

73% vs 23%

Agent turns

0%

1.0 vs 1.0

Tokens

-17%

1,365.83 vs 1,639.5

fail → pass · 12pass → pass · 4fail → fail · 5pass → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-12 What JSON key in the indexer summary report tracks relationships created between code elements?	fail→pass	10,897	1,826	-83%	1	1	0%	1,690	872	-48%	0	0	—
case-01 I just pulled the latest changes in our repository. Can you index the codebase to map out our code knowledge graph and give me a summary of how many files, nodes, and edges were processed along with the elapsed time and suggested follow-up tasks?	fail→fail	8,205	6,667	-19%	1	1	0%	1,183	928	-22%	0	0	—
case-02 Please generate the structural graph representation for this workspace directory. When it finishes, display the summary JSON metrics for files scanned, nodes, edges, and duration, then let me know what analysis steps I should perform next.	fail→fail	8,244	6,922	-16%	1	1	0%	1,471	977	-34%	0	0	—
case-03 I want to prepare our project for flow tracing and architectural analysis. Could you build or refresh the codebase knowledge database for the current project folder and print out the build stats report alongside recommended next actions?	fail→fail	2,595	6,085	+134%	1	1	0%	255	903	+254%	0	0	—
case-04 I already built the code structure database earlier today. Can you search the database to list all functions that call the user authentication handler in src/auth.ts?	fail→pass	4,471	24,698	+452%	1	1	0%	640	4,084	+538%	0	0	—
case-05 We need to extract the challenge knowledge base artifacts from our benchmark evaluation suite.	fail→pass	12,833	8,924	-30%	1	1	0%	2,320	2,076	-11%	0	0	—
case-06 Our SQLite graph file is already built and up to date. Can you query it to show all classes inheriting from BaseController?	fail→fail	6,571	10,402	+58%	1	1	0%	1,183	965	-18%	0	0	—
case-07 I made minor edits to two files in src/ core module. The file .gauntlet/graph.db already exists. What exact command should be executed to refresh the index?	fail→pass	5,481	3,024	-45%	1	1	0%	907	1,097	+21%	0	0	—
case-08 This is a fresh repository clone and no .gauntlet directory exists yet. What script command should be run to build the initial structural map?	fail→pass	7,115	4,190	-41%	1	1	0%	1,116	1,306	+17%	0	0	—
case-09 When creating the codebase SQLite graph at .gauntlet/graph.db, how should git tracking be handled so binary database files are not committed?	pass→pass	9,277	3,891	-58%	1	1	0%	1,555	1,234	-21%	0	0	—
case-10 I am writing a script runner for index creation. What specific JSON key name must hold the count of scanned source files in the execution summary?	fail→pass	9,251	1,866	-80%	1	1	0%	1,383	879	-36%	0	0	—
case-11 What JSON field name in the indexing build report represents the total graph nodes created?	pass→pass	10,506	1,644	-84%	1	1	0%	1,663	831	-50%	0	0	—
case-13 Which key name in the indexing execution JSON report measures the total elapsed build time?	fail→pass	12,421	1,401	-89%	1	1	0%	1,940	768	-60%	0	0	—
case-14 We are inspecting the backend database file created at .gauntlet/graph.db. What database engine and journal mode are configured for this file?	pass→fail	6,257	4,363	-30%	1	1	0%	1,026	898	-12%	0	0	—
case-15 When doing incremental updates on modified source files, how does the indexer detect which files remained unchanged without re-parsing everything?	fail→pass	14,347	4,567	-68%	1	1	0%	2,352	1,301	-45%	0	0	—
case-16 What structural element types are extracted as nodes when parsing AST structures across supported source files?	fail→pass	16,228	3,374	-79%	1	1	0%	2,815	1,064	-62%	0	0	—
case-17 What relationship edge types are mapped between symbols during AST index construction?	fail→pass	12,707	3,010	-76%	1	1	0%	2,048	1,108	-46%	0	0	—
case-18 We want to index a subproject located specifically at ./packages/backend rather than the root directory. How is this target passed to the builder script?	fail→pass	10,624	3,062	-71%	1	1	0%	1,843	1,102	-40%	0	0	—
case-19 After completing a graph build for a refactoring session, what specific impact analysis capability should be suggested to the user?	pass→pass	9,338	2,195	-76%	1	1	0%	1,443	881	-39%	0	0	—
case-20 Once the initial indexing build finishes, what is the primary recommended follow-up action for exploring symbol relationships?	pass→pass	8,614	2,356	-73%	1	1	0%	1,295	965	-25%	0	0	—
case-21 Does the AST indexing tool support multi-language codebases containing Go, Rust, Java, C/C++, TypeScript, and Python?	fail→pass	11,237	3,965	-65%	1	1	0%	1,900	1,298	-32%	0	0	—
case-22 We just initialized a new AI pair programming session on a complex unfamiliar repository. What action should be performed to establish structural awareness?	fail→fail	9,553	11,586	+21%	1	1	0%	1,575	2,821	+79%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 22.7%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 18
Cases that regressed: 2
Skill version measured: v1
Run completed: 8/20/2026, 22:40 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-12

Prompt

What JSON key in the indexer summary report tracks relationships created between code elements?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signalsno signals yet

BenchmarkRun a verified benchmark with positive lift

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

No signals yet — run a benchmark or collect evaluated traces to earn a score.

Verified on gemini-3.6-flash · re-tested today

Other measured skills in the registry, with their headline benchmark lift.