Summary

.tf, .tfvars, and .hcl files are skipped by gbrain sync --strategy code because none of the HCL extensions are in CODE_EXTENSIONS at src/core/sync.ts:46-75. Terraform repos (and any other HCL-based infra tooling — Packer, Vault, Nomad, Consul) have the vast majority of their app surface in .tf / .tfvars; gbrain currently treats them as invisible.

Repro

$ find . -name '*.tf' ! -path '*/.terraform/*' | wc -l
163
$ find . -name '*.tfvars' ! -path '*/.terraform/*' | wc -l
51

$ gbrain sync --source <id> --strategy code
[gbrain phase] import.collect_files done 4ms files=6
Found 6 code files                # 0 .tf, 0 .tfvars
…
$ gbrain sources list --json | jq '.sources[] | select(.id=="<id>") | .page_count'
7                                  # 6 .json + 1 .py only — entire IaC body skipped

Root cause

src/core/sync.ts:46-75 declares the allowlist:

const CODE_EXTENSIONS = new Set<string>([
  '.ts', '.tsx', '.mts', '.cts',
  '.js', '.jsx', '.mjs', '.cjs',
  '.py', '.rb', '.go', '.rs', '.java', '.cs',
  '.cpp', '.cc', '.cxx', '.hpp', '.hxx', '.hh',
  '.c', '.h',
  '.php', '.swift', '.kt', '.kts', '.scala', '.sc',
  '.lua', '.ex', '.exs', '.elm', '.ml', '.mli',
  '.dart', '.zig', '.sol',
  '.sh', '.bash',
  '.css', '.html', '.htm',
  '.vue',
  '.json', '.yaml', '.yml', '.toml',
]);

Note .toml and .yaml are included as config-file extensions. .tf / .tfvars are the structural equivalent for Terraform: a declarative config language with named blocks, variables, resource references, and module imports — same indexability profile as YAML/TOML and same need for code-aware chunking.

The comment at line 35-44 says this allowlist is "kept as-is for now for isAllowedByStrategy fast-path + tests" while detectCodeLanguage is the long-term source of truth — so HCL may also need adding to whatever map detectCodeLanguage uses, depending on which one wins at runtime.

Suggested fix

Add HCL extensions to CODE_EXTENSIONS:

'.toml', '.hcl', '.tf', '.tfvars',

A first-party tree-sitter grammar exists (tree-sitter-hcl) and covers both .tf and .hcl. If the chunker has no HCL parser yet, the existing config-style fallback that handles .toml should chunk .tf / .tfvars reasonably — they're block-structured and line-oriented, well-suited to the recursive chunker.

Why it matters

Terraform is the dominant IaC tool in the Cloudflare / AWS / Azure ecosystem, and HCL is also the config language for Packer, Vault, Nomad, Consul, and Waypoint. A typical infra repo is 80-95% .tf / .tfvars by file count. Without HCL support, gbrain code-def / code-refs / code-callers return empty for every module, resource, and variable in the repo — exactly the questions an SRE / platform engineer would want to ask the brain ("where is azurerm_kubernetes_cluster configured?", "what calls module aks?", "which env declares enable_auto_upgrade?").

This is the same pattern as #709 (.astro missing), just for the IaC ecosystem instead of the static-site one.

Environment

• gbrain: v0.31.3 (commit 9c60b3a, master)
• Bun: 1.3.11
• Database: Akamai managed Postgres
• Platform: Linux 6.8.0-111-generic x86_64
• Repo: 163 .tf + 51 .tfvars + 20 .md + 9 .json — code sync indexed 7 of the 243 files

Related

• bug: .astro extension missing from CODE_EXTENSIONS — Astro projects' SFCs invisible to code sync #709 (.astro missing) — exact same shape, different ecosystem. Filing separately per the maintainer's labeling cadence; this one is small enough to be a good-first-issue candidate.
• No recovery path when initial code-source import fails to embed (e.g. missing OPENAI_API_KEY at first sync) #877 (no recovery path when initial code-source import fails to embed) — adjacent reliability issue surfaced while reproducing this one.
• bug: code indexing doesn't work as expected (or at all) #457 (source-routing + --strategy code ignored) — broader code-indexing umbrella.