diff --git a/.github/ISSUE_TEMPLATE/config.yml b/.github/ISSUE_TEMPLATE/config.yml
new file mode 100644
index 0000000..8cd084d
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/config.yml
@@ -0,0 +1,10 @@
+blank_issues_enabled: true
+contact_links:
+  # TODO(maintainer): replace the OWASP Slack invite URL below with the direct
+  # invite link to the initiative's channel (e.g. #genai-data-security) once known.
+  - name: OWASP GenAI Slack (DSGAI scanner discussion)
+    url: https://owasp.org/slack/invite
+    about: >-
+      Questions, scan reports, or general discussion? Join the OWASP Slack and find
+      the GenAI Data Security Initiative channel. For wrong scanner results, please
+      still file a [scanner] false-positive / false-negative issue so it becomes a test case.
diff --git a/.github/ISSUE_TEMPLATE/scanner-bug.yml b/.github/ISSUE_TEMPLATE/scanner-bug.yml
new file mode 100644
index 0000000..c869a6f
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/scanner-bug.yml
@@ -0,0 +1,53 @@
+name: "[scanner] Bug"
+description: A general behavior bug in the DSGAI scanner (report rendering, CI, flags, CLI).
+title: "[scanner] Bug: "
+labels: ["scanner", "bug"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        For wrong findings, prefer the false-positive / false-negative templates.
+        Use this one for everything else: crashes, rendering, CI, flag handling.
+  - type: textarea
+    id: what-happened
+    attributes:
+      label: What happened
+      description: Describe the bug and what you expected instead.
+    validations:
+      required: true
+  - type: textarea
+    id: repro
+    attributes:
+      label: Steps to reproduce
+      description: Exact commands / actions. MUST NOT contain real secrets.
+      render: text
+    validations:
+      required: true
+  - type: dropdown
+    id: area
+    attributes:
+      label: Area
+      options:
+        - Report rendering (HTML)
+        - CLI (dsgai_scan.py)
+        - Claude Code skill
+        - GitHub Action / CI
+        - Flags / arguments
+        - Other
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Skill / CLI version
+      placeholder: "0.3.0"
+    validations:
+      required: true
+  - type: textarea
+    id: context
+    attributes:
+      label: Logs / additional context
+      description: Paste relevant output. Redact anything sensitive.
+      render: text
+    validations:
+      required: false
diff --git a/.github/ISSUE_TEMPLATE/scanner-false-negative.yml b/.github/ISSUE_TEMPLATE/scanner-false-negative.yml
new file mode 100644
index 0000000..5658bb0
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/scanner-false-negative.yml
@@ -0,0 +1,56 @@
+name: "[scanner] False negative"
+description: The DSGAI scanner missed something it should have flagged.
+title: "[scanner] False negative: "
+labels: ["scanner", "false-negative"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for improving the DSGAI scanner's coverage. Every accepted false-negative
+        report becomes a permanent positive test case in the public fixture corpus, with credit.
+  - type: textarea
+    id: what-missed
+    attributes:
+      label: What should have been flagged
+      description: Describe the issue the scanner failed to detect.
+    validations:
+      required: true
+  - type: textarea
+    id: snippet
+    attributes:
+      label: Minimal reproducing snippet
+      description: The smallest code that should have triggered a finding. MUST NOT contain real secrets — use obviously fake values.
+      render: text
+    validations:
+      required: true
+  - type: input
+    id: expected-control
+    attributes:
+      label: Expected control
+      description: Which DSGAI control should this map to, e.g. `DSGAI02`.
+      placeholder: DSGAI02
+    validations:
+      required: true
+  - type: dropdown
+    id: expected-severity
+    attributes:
+      label: Expected severity
+      options:
+        - FAIL
+        - WARN
+        - PASS signal
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Skill / CLI version
+      description: e.g. `0.3.0`. See the report header or `dsgai_scan.py --version`.
+      placeholder: "0.3.0"
+    validations:
+      required: true
+  - type: markdown
+    attributes:
+      value: |
+        > Reproducing snippets are added to the public test fixture corpus under the repo license.
+        > Never paste real credentials — use obviously fake values.
diff --git a/.github/ISSUE_TEMPLATE/scanner-false-positive.yml b/.github/ISSUE_TEMPLATE/scanner-false-positive.yml
new file mode 100644
index 0000000..2e2e353
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/scanner-false-positive.yml
@@ -0,0 +1,73 @@
+name: "[scanner] False positive"
+description: The DSGAI scanner flagged something that is not actually a problem.
+title: "[scanner] False positive: "
+labels: ["scanner", "false-positive"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for improving the DSGAI scanner's precision. Every accepted false-positive
+        report becomes a permanent negative test case in the public fixture corpus, with credit.
+  - type: input
+    id: pattern-id
+    attributes:
+      label: Pattern ID
+      description: The rule that fired, e.g. `P12.1`. Find it in the report next to the finding.
+      placeholder: P12.1
+    validations:
+      required: true
+  - type: input
+    id: control
+    attributes:
+      label: DSGAI control
+      description: The control the rule maps to, e.g. `DSGAI12`.
+      placeholder: DSGAI12
+    validations:
+      required: false
+  - type: textarea
+    id: snippet
+    attributes:
+      label: Minimal reproducing snippet
+      description: The smallest code that triggers the false positive. MUST NOT contain real secrets — use obviously fake values.
+      render: text
+    validations:
+      required: true
+  - type: textarea
+    id: reported
+    attributes:
+      label: What the scanner reported
+      description: Paste the finding line(s) from the report or SARIF.
+    validations:
+      required: true
+  - type: textarea
+    id: why-wrong
+    attributes:
+      label: Why it's wrong
+      description: Explain why this code is safe / why the finding is incorrect.
+    validations:
+      required: true
+  - type: dropdown
+    id: environment
+    attributes:
+      label: Environment
+      description: How did you run the scanner?
+      options:
+        - Claude Code skill (dsgai_scanner_tool.md)
+        - Deterministic CLI (cli/dsgai_scan.py)
+        - GitHub Action
+        - Other tool (prompt variant)
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Skill / CLI version
+      description: e.g. `0.3.0`. See the report header or `dsgai_scan.py --version`.
+      placeholder: "0.3.0"
+    validations:
+      required: true
+  - type: markdown
+    attributes:
+      value: |
+        > Reproducing snippets are added to the public test fixture corpus under the repo license.
+        > Never paste real credentials — use obviously fake values.
diff --git a/.github/ISSUE_TEMPLATE/scanner-new-rule.yml b/.github/ISSUE_TEMPLATE/scanner-new-rule.yml
new file mode 100644
index 0000000..2b9eeaa
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/scanner-new-rule.yml
@@ -0,0 +1,58 @@
+name: "[scanner] Propose a new rule"
+description: Propose a new detection rule for a DSGAI control.
+title: "[scanner] New rule: "
+labels: ["scanner", "rules"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for extending DSGAI coverage. Accepted rules ship with your test snippets
+        as permanent fixture cases, with credit. Rules cannot be merged without at least
+        one positive and one negative test case.
+  - type: input
+    id: control
+    attributes:
+      label: Proposed control mapping
+      description: Which DSGAI control does this rule serve, e.g. `DSGAI17`.
+      placeholder: DSGAI17
+    validations:
+      required: true
+  - type: textarea
+    id: pcre
+    attributes:
+      label: Proposed PCRE pattern
+      description: The regex (PCRE2, ripgrep-compatible). Give it as a code block, not a table cell.
+      render: text
+    validations:
+      required: true
+  - type: dropdown
+    id: classification
+    attributes:
+      label: Classification
+      description: STRUCTURAL rules match code shape; VALUE-BEARING rules match around secret material and run in location-only mode.
+      options:
+        - STRUCTURAL
+        - VALUE-BEARING
+    validations:
+      required: true
+  - type: textarea
+    id: positive
+    attributes:
+      label: Positive test snippet(s)
+      description: Code that MUST match. Fake values only. Goes straight into the fixture app on acceptance.
+      render: text
+    validations:
+      required: true
+  - type: textarea
+    id: negative
+    attributes:
+      label: Negative test snippet(s)
+      description: Similar-looking code that MUST NOT match. Fake values only.
+      render: text
+    validations:
+      required: true
+  - type: markdown
+    attributes:
+      value: |
+        > Test snippets are added to the public test fixture corpus under the repo license.
+        > Never paste real credentials — use obviously fake values.
diff --git a/dsgai_scanner_tool/CHANGES_v0.3.md b/dsgai_scanner_tool/CHANGES_v0.3.md
new file mode 100644
index 0000000..83867a2
--- /dev/null
+++ b/dsgai_scanner_tool/CHANGES_v0.3.md
@@ -0,0 +1,20 @@
+# Changelog — DSGAI Scanner Tool v0.3
+
+All notable changes for the v0.3 line. This file is appended by every merged PR.
+v0.3 releases when Phases 0–2 of the improvement plan are complete.
+
+Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/) loosely;
+dates are ISO-8601. The previous line is recorded in [`CHANGES_v0.2.md`](CHANGES_v0.2.md).
+
+## [Unreleased]
+
+### Added
+- Contributor infrastructure: `[scanner]` GitHub issue-form templates (false-positive,
+  false-negative, new-rule, bug), scanner `CONTRIBUTING.md`, public `ROADMAP.md`, and
+  this changelog scaffold. (PR-01)
+
+### Changed
+- _nothing yet_
+
+### Fixed
+- _nothing yet_
diff --git a/dsgai_scanner_tool/CONTRIBUTING.md b/dsgai_scanner_tool/CONTRIBUTING.md
new file mode 100644
index 0000000..09615f8
--- /dev/null
+++ b/dsgai_scanner_tool/CONTRIBUTING.md
@@ -0,0 +1,94 @@
+# Contributing to the DSGAI Scanner Tool
+
+The DSGAI scanner is a Claude Code skill (plus a deterministic CLI, landing across
+v0.3) that audits GenAI applications against the 21 [OWASP GenAI Data Security
+controls](https://genai.owasp.org/resource/owasp-genai-data-security-risks-mitigations-2026/).
+It gets more useful every time someone runs it on a real repo and tells us what it
+got wrong. That is the single most valuable thing you can do here.
+
+> This file covers the **scanner subproject** (`dsgai_scanner_tool/`). For the wider
+> initiative, see the [repository-root CONTRIBUTING](../CONTRIBUTING.md).
+
+## Fastest way to help
+
+1. Run the scan on one of your own GenAI repos.
+2. When it flags something wrong, or misses something real, open an issue with the
+   [scanner templates](../.github/ISSUE_TEMPLATE/):
+   - **[scanner] False positive** — it flagged safe code.
+   - **[scanner] False negative** — it missed a real problem.
+   - **[scanner] Propose a new rule** — a control needs coverage it doesn't have.
+   - **[scanner] Bug** — anything else (rendering, CLI, CI, flags).
+
+**Every accepted false-positive / false-negative report becomes a permanent test case
+in the public fixture corpus, credited to you.** You don't have to write any code to
+make the tool measurably better — a good repro is the contribution.
+
+> Never paste real credentials into an issue or a fixture. Use obviously fake,
+> canonical values (e.g. `sk-proj-FAKE00000000000000000000000000`).
+
+## Contributing a rule
+
+Detection rules are moving from prose in `dsgai_scanner_tool.md` into data at
+`rules/dsgai-rules.yaml`, validated by `rules/rules.schema.json`. **(Landing soon —
+tracked by PR-03.)** Until that lands, describe your rule in a
+[new-rule issue](../.github/ISSUE_TEMPLATE/scanner-new-rule.yml) using the format below;
+once the YAML rule format ships, this section will point at `rules/README.md`.
+
+Every rule needs:
+
+- **A control mapping** — one of `DSGAI01`–`DSGAI21`.
+- **A PCRE pattern** — PCRE2 syntax, runs under `rg --pcre2`. Give it in a fenced code
+  block, never a markdown table cell (`\|` becomes a literal pipe when the file is read
+  back and silently breaks alternation).
+- **A classification** (see below).
+- **At least one positive and one negative test snippet.** No exceptions — a rule
+  without a negative case has no defined precision.
+
+### STRUCTURAL vs VALUE-BEARING
+
+This classification controls *how* a rule executes and *what may appear in a report*.
+
+| Class | The match is… | Report may show the matched line? | Execution |
+|---|---|---|---|
+| **STRUCTURAL** | a code shape — a call, a missing import, a missing decorator, an architectural gap. Contains no runtime secret. | Yes | normal `rg` match |
+| **VALUE-BEARING ⚠️** | a line whose *content is* a credential, key, secret, connection string, or PII. | **Never** — not in the report, checkpoint, or any persisted tool call, even if the value looks fake. | location-only: `rg -n -o --replace '' --pcre2` so ripgrep erases the match before emitting `path:line:` |
+
+The four VALUE-BEARING controls today are **DSGAI02, DSGAI13, DSGAI14, DSGAI15**. If
+your rule matches lines that could contain secret material, it is VALUE-BEARING — when
+in doubt, classify up.
+
+### Running the self-test locally
+
+Once the test infrastructure lands (PR-04/PR-05/PR-06):
+
+```bash
+pip install pyyaml pytest
+pytest dsgai_scanner_tool/tests/ -q
+```
+
+The suite compiles every rule's PCRE, scans the fixture app, and checks the result
+against `tests/expected-findings.yaml`. **Rule PRs are only safely reviewable once this
+infrastructure exists** — if you draft a rule PR before then, that's welcome, just
+expect it to sit until the fixtures land (see the `blocked-on-phase-1` label).
+
+## Code contributions
+
+- **Branch naming:** `improve/pr-NN-short-slug` for planned work, or a short descriptive
+  slug otherwise. Never commit to `main`.
+- **Conventional commits:** `fix:`, `feat:`, `docs:`, `ci:`, `test:`, `refactor:`.
+- **Shell scripts** must be `shellcheck` clean and portable (bash 3.2 / macOS included —
+  no `mapfile`, `readarray`, `${var,,}`, or GNU-only `grep` flags).
+- **CI must be green** — the scanner has path-filtered lint and self-test workflows.
+
+## What we will NOT merge
+
+The scanner has deliberate limits that keep it maintainable and trustworthy. See the
+[Non-goals section of the README](README.md#non-goals) — please read it before
+proposing large additions, so your effort lands somewhere we can accept it.
+
+## Recognition
+
+Contributors are credited by handle in every release note. **Recurring contributors are
+invited to become co-maintainers** — this project explicitly wants to grow its
+maintainer base, and consistent, high-quality contributors are how that happens. If you
+land a few accepted issues or PRs, expect an invitation.
diff --git a/dsgai_scanner_tool/ROADMAP.md b/dsgai_scanner_tool/ROADMAP.md
new file mode 100644
index 0000000..1632360
--- /dev/null
+++ b/dsgai_scanner_tool/ROADMAP.md
@@ -0,0 +1,68 @@
+# DSGAI Scanner Tool — Roadmap
+
+This is the public roadmap for the DSGAI scanner. It turns the internal improvement
+plan into work anyone can pick up.
+
+> **Want one of these? Comment on the tracking issue and claim it.** Each unstarted
+> item below links to a GitHub issue. Rule and precision work is labelled
+> `blocked-on-phase-1` until the test corpus and self-test CI land — you can draft it
+> now, it just can't merge until then.
+
+Status legend: ✅ done · 🚧 in progress · ⬜ not started
+
+## Phase 1 — Trust foundation (determinism, tests, schemas)
+
+A compliance report that changes run-to-run on identical input is an opinion, not
+evidence. This phase makes pattern matching deterministic and testable, so everything
+after it is verifiable.
+
+- ⬜ **Rules as data** — extract every detection pattern into `rules/dsgai-rules.yaml`
+  with a JSON Schema, so rules are reviewable data instead of prose.
+- ⬜ **Fixture app + known-answer sheet** — a small, intentionally vulnerable
+  multi-language GenAI app that is the test corpus, the demo, and the contributor
+  on-ramp.
+- ⬜ **Deterministic runner** — a stdlib Python CLI that runs the rules via ripgrep and
+  emits findings JSON + SARIF, identically every run.
+- ⬜ **Checkpoint schema + self-test CI** — a formal, redaction-checkable output schema
+  and the CI gate that makes external rule PRs safe to merge.
+
+## Phase 2 — Secure the pipeline itself
+
+A security scanner that is itself a prompt-injection vector or a secret-leak channel is
+a liability. This phase closes those.
+
+- ⬜ **Skill rewrite** — the LLM becomes the orchestrator; the deterministic engine owns
+  pattern matching. Adds an untrusted-content trust preamble, structural secret
+  redaction, stable file IDs, and honest language about what the report guarantees.
+- ⬜ **Harden the GitHub Action** — split scanning (no secrets, runs on forks) from
+  narration (restricted tools), pin actions by SHA, and fix the push-gate behaviour.
+- ⬜ **README truth pass + lighter sample image** — every claim matches reality; the
+  ~5 MB sample screenshot is replaced with a small one from the public fixture app.
+- ⬜ **Pre-commit: gitleaks rule pack** — ship a battle-tested gitleaks pack, keep a
+  portable no-dependency fallback script.
+
+## Phase 3 — Precision
+
+- ⬜ **Pattern precision wave 1** — fix confirmed false positives/negatives (unquoted
+  `.env` keys, the innocent-webhook SQL false positive), add per-rule confidence levels.
+- ⬜ **CVE pipeline rework** — move CVE fetching into the CLI (no hallucinated CVEs),
+  add caching, inline suppressions with reasons, a baseline for CI gating, and
+  incremental `--diff` scans.
+
+## Phase 4 — Professional polish
+
+- ⬜ **License split + SPDX headers** — content stays CC BY-SA 4.0; executable code
+  moves to Apache-2.0 (pending OWASP leadership sign-off).
+- ⬜ **Single-source variants + templated report + static ATLAS map** — generate the
+  tool-neutral prompt variant from the skill, render reports deterministically from a
+  template, and ship a static MITRE ATLAS technique map.
+- ⬜ **Ecosystem expansion** — C#/NuGet, Rust, Ruby coverage, plus Semgrep and gitleaks
+  rule-pack exports so incumbent toolchains carry the DSGAI framework.
+- ⬜ **Benchmark + published precision report** — run against public GenAI repos,
+  hand-label, and publish per-rule precision. This is what separates "interesting
+  project" from "tool AppSec teams adopt".
+
+---
+
+*The roadmap follows the initiative's improvement plan. Checkboxes are updated as work
+merges; each item's tracking issue carries the detail and the discussion.*