---
phase: 01-foundation
plan: 02
subsystem: providers
tags: [yaml, embed, aho-corasick, registry, go-embed, gopkg.in/yaml.v3]

# Dependency graph
requires:
  - phase: 01-01
    provides: go.mod with all Phase 1 dependencies, test scaffolding, cmd/root.go stub
provides:
  - Provider, Pattern, VerifySpec, RegistryStats Go structs with YAML validation
  - Registry with List(), Get(), Stats(), AC() methods
  - Aho-Corasick automaton built from all provider keywords at NewRegistry()
  - Three reference provider YAML definitions (openai, anthropic, huggingface)
  - Compile-time embed of provider YAML via pkg/providers/definitions/
affects:
  - scan-engine
  - cli-providers-command
  - verification-engine
  - storage-layer

# Tech tracking
tech-stack:
  added:
    - gopkg.in/yaml.v3 (UnmarshalYAML custom validation)
    - github.com/petar-dambovaliev/aho-corasick (keyword pre-filter automaton)
    - embed (stdlib) for compile-time YAML embedding
  patterns:
    - Provider YAML at providers/ (user-visible) + pkg/providers/definitions/ (embed location)
    - Type alias pattern for custom UnmarshalYAML without infinite recursion
    - Registry injected via constructor (NewRegistry), not global singleton

key-files:
  created:
    - pkg/providers/schema.go
    - pkg/providers/loader.go
    - pkg/providers/registry.go
    - pkg/providers/registry_test.go
    - pkg/providers/definitions/openai.yaml
    - pkg/providers/definitions/anthropic.yaml
    - pkg/providers/definitions/huggingface.yaml
    - providers/openai.yaml
    - providers/anthropic.yaml
    - providers/huggingface.yaml
  modified: []

key-decisions:
  - "Provider YAML kept in dual locations: providers/ (user-visible) and pkg/providers/definitions/ (embedded) — Go embed cannot use '..' paths, so definitions/ subdirectory is canonical embed source"
  - "UnmarshalYAML validates format_version >= 1 and non-empty last_verified at parse time, not at registry use time — fail fast on malformed definitions"
  - "Aho-Corasick automaton built with DFA=true for deterministic performance — trades memory for guaranteed O(n) matching"
  - "Registry is value-safe for concurrent reads — no mutex needed since providers slice is written once at NewRegistry and never mutated"

patterns-established:
  - "Pattern 1: Type alias in UnmarshalYAML to avoid infinite recursion: `type ProviderAlias Provider`"
  - "Pattern 2: embed path convention — YAML at pkg/providers/definitions/, user docs at providers/"
  - "Pattern 3: Registry constructor NewRegistry() loads+validates+indexes+builds AC in one call"

requirements-completed: [CORE-02, CORE-03, CORE-06, PROV-10]

# Metrics
duration: 9min
completed: 2026-04-04
---

# Phase 01 Plan 02: Provider Registry Summary

**YAML schema structs with UnmarshalYAML validation, embed.FS loader, and Aho-Corasick registry serving List/Get/Stats/AC to all downstream subsystems**

## Performance

- **Duration:** ~9 min
- **Started:** 2026-04-04T21:02:31Z
- **Completed:** 2026-04-04T21:11:41Z
- **Tasks:** 2 (both TDD)
- **Files modified:** 10 created, 1 updated (registry_test.go)

## Accomplishments

- Provider YAML schema with compile-time validation (format_version >= 1, last_verified required, confidence enum)
- Registry loads 3 providers from embedded YAML at startup, builds Aho-Corasick automaton over all keywords
- Three reference provider YAML definitions with full verify specs (OpenAI, Anthropic, HuggingFace)
- All 5 provider tests pass: TestRegistryLoad, TestRegistryGet, TestRegistryStats, TestAhoCorasickBuild, TestProviderSchemaValidation

## Task Commits

Each task was committed atomically:

1. **TDD RED - Failing tests for schema and registry** - `ebaf7d7` (test)
2. **Task 1: Provider schema structs and reference YAMLs** - `4fcdc42` (feat)
3. **Task 2: Embed loader, registry with AC, filled test stubs** - `a9859b3` (feat)

_Note: Bootstrap (go.mod, main.go, test stubs) was included in the RED commit since Plan 01-01 runs in parallel._

## Files Created/Modified

- `pkg/providers/schema.go` - Provider, Pattern, VerifySpec, RegistryStats structs with UnmarshalYAML validation
- `pkg/providers/loader.go` - embed.FS declaration with //go:embed definitions/*.yaml and fs.WalkDir loader
- `pkg/providers/registry.go` - Registry struct with List(), Get(), Stats(), AC() methods and NewRegistry() constructor
- `pkg/providers/registry_test.go` - Full test implementation (replaced stub from Plan 01)
- `pkg/providers/definitions/openai.yaml` - Embedded OpenAI provider definition
- `pkg/providers/definitions/anthropic.yaml` - Embedded Anthropic provider definition
- `pkg/providers/definitions/huggingface.yaml` - Embedded HuggingFace provider definition
- `providers/openai.yaml` - User-visible OpenAI reference definition
- `providers/anthropic.yaml` - User-visible Anthropic reference definition
- `providers/huggingface.yaml` - User-visible HuggingFace reference definition

## Decisions Made

- **Dual YAML location:** providers/ for user reference, pkg/providers/definitions/ for embed — Go's embed package cannot traverse `..` paths, so definitions/ inside the package is the only valid embed location.
- **DFA mode for Aho-Corasick:** `Opts{DFA: true}` chosen for guaranteed O(n) matching at cost of higher upfront build time — appropriate for a scanner tool that pays build cost once and scans many files.
- **Constructor injection over globals:** NewRegistry() returns a value; callers inject it. No package-level `var Registry` global — avoids init order issues and enables testing.

## Deviations from Plan

### Auto-fixed Issues

**1. [Rule 3 - Blocking] Bootstrapped Plan 01-01 prerequisites in this worktree**
- **Found during:** Pre-task setup
- **Issue:** Plan 01-02 depends on Plan 01-01 (go.mod, main.go, test stubs) which runs in parallel in a different worktree. This worktree had no go.mod.
- **Fix:** Executed Plan 01-01 bootstrap (go mod init, go get all 10 deps, main.go, cmd/root.go, testdata fixtures, test stub files) before starting Plan 01-02 tasks.
- **Files modified:** go.mod, go.sum, main.go, cmd/root.go, testdata/samples/*.txt, pkg/*/stub_test.go files
- **Verification:** `go build ./...` succeeded before Plan 01-02 task execution
- **Committed in:** ebaf7d7 (RED phase commit includes bootstrap)

**2. [Rule 3 - Blocking] go mod tidy required after adding production packages**
- **Found during:** Task 2 GREEN phase
- **Issue:** `go test` failed with "no required module provides package github.com/petar-dambovaliev/aho-corasick" even though it was in go.mod — tidy hadn't propagated it for non-test code.
- **Fix:** Ran `go mod tidy` which resolved the module graph.
- **Files modified:** go.mod, go.sum
- **Verification:** `go test ./pkg/providers/...` passed after tidy

---

**Total deviations:** 2 auto-fixed (2 blocking/infrastructure)
**Impact on plan:** Both deviations were infrastructure setup, not scope changes. Plan objectives met exactly.

## Issues Encountered

- Go embed `..` path restriction required dual YAML directory strategy (documented in plan's context, confirmed during implementation)
- aho-corasick package name is `aho_corasick` (underscore) not `ahocorasick` — used import alias `ahocorasick` for cleaner code

## User Setup Required

None - no external service configuration required.

## Next Phase Readiness

- Registry interface is stable: NewRegistry(), List(), Get(), Stats(), AC() — downstream plans can depend on these signatures
- Phase 03 (Storage Layer) can proceed immediately — no registry dependency
- Phase 04 (Scan Engine) can now wire AC() for keyword pre-filtering
- Phase 05 (CLI) can call Registry.List() for `keyhunter providers list`
- Known: only 3 reference providers embedded; Phase 02-03 will add all 108

---
*Phase: 01-foundation*
*Completed: 2026-04-04*