securable-claude-plugin

A Claude Code plugin offering secure code generation and securability analysis through application of the OWASP FIASSE: The Securable framework. Also, part of OWASP Secure Agent Playbook.

Overview

This plugin augments Claude Code with three capabilities:

Securability Engineering Review — Analyze existing code for securable qualities using the ten SSEM attributes (FIASSE v1.1) across three pillars (Maintainability, Trustworthiness, Reliability), producing scored assessments with actionable findings.
Securability Engineering Code Generation — Generate new code that embodies securable qualities by default, applying OWASP FIASSE principles as engineering constraints.
PRD Securability Enhancement — Enhance product requirements documents with ASVS level selection, feature-level ASVS requirement mapping, SSEM implementation annotations, and FIASSE tenet coverage.
FIASSE Lookup — Answer FIASSE/SSEM questions (definitions, principles, attributes, scoring conduct, section numbers) from the bundled framework reference, citing section numbers.

Agent-facing guidance lives in AGENTS.md (the AGENTS.md standard entry point, imported by CLAUDE.md). A critical assessment of this plugin and its enhancement roadmap lives in docs/critical-review-2026-08.md.

Elevating any harness

Beyond the skills, the pack carries its impact in three layers so any code-generation harness — not only the one it's loaded into — is bound by it:

The securability kernel (core/kernel.md) — a ~300-token always-on distillation of the non-negotiables (parse-don't-trust, server-side authority, the never-emit list, observable security, Securability Notes). scripts/build_bindings.py generates per-harness bindings from it — Cursor rule, Copilot instructions, Gemini CLI context, Aider conventions, and the kernel block inside AGENTS.md — with a size budget and a CI drift guard (--check). Bindings are generated, never edited. Measured (kernel A/B, tests/kernel_ab.py, deterministic detectors, claude CLI): on a strong frontier model the baseline already avoided all detector anti-patterns (0/0 — verified real, not detector blindness), while the kernel's process contract was adopted 3/3 (Securability Notes present) vs 0/3 baseline. Anti-pattern deltas are expected to show on weaker models — measuring that per harness is the cross-harness scoreboard's job (roadmap M5).
The securable contract (.securable/requirements.yaml + boundaries.yaml in consuming projects) — repo-resident, machine-readable requirements with testable acceptance criteria and a planned → implemented → verified lifecycle. Emitted by the PRD skill, honored by generation in any harness, verified at review. scripts/validate_securable.py enforces the semantics — including that every cited ASVS ID exists in the bundled 5.0 catalog. See docs/securable-contract.md.
Held checks — rules/opengrep/securable.yaml maps the skills' anti-pattern tags to enforceable opengrep rules (SSEM/ASVS metadata on each; opengrep is the LGPL, community-governed scanner), tested against paired fail/pass fixtures; and scripts/securability_report.sh + .github/workflows/securability-report.yml produce the FIASSE S5.2.1 Securability Report on pull requests via any agent CLI (claude, codex, opencode) — advisory by default, per S5.2.2.

Run everything CI runs with scripts/run_checks.sh.

Installation

One skills tree, one adapter per agent — the multi-agent packaging methodology of obra/superpowers. Each supported harness gets a root-level adapter (a manifest in that harness's plugin format, or an agent-followable INSTALL.md), and every adapter points at the same skills/ + data/ tree. scripts/check_manifests.py (run by CI) keeps the manifests in lockstep.

Agent	Adapter	Install
Claude Code	`.claude-plugin/`	`/plugin` marketplace
Cursor	`.cursor-plugin/`	plugin marketplace, or kernel rule + skills copy
Devin	`.devin-plugin/`	`devin plugins install Securability-Engineering/securable-claude-plugin`
opencode	`.opencode/INSTALL.md`	fetch-and-follow
Any AGENTS.md/SKILL.md agent	`.agents/INSTALL.md`	fetch-and-follow, or `scripts/install_skills.sh`

Claude Code

Install through the Claude Code plugin manager: open the interactive manager with /plugin, then use the Discover and Marketplaces tabs to add/install graphically. Or from this repository's marketplace manifest:

/plugin marketplace add Securability-Engineering/securable-claude-plugin
/plugin install securable-claude-plugin@securable-claude-plugins

Cursor

.cursor-plugin/plugin.json declares the pack in Cursor's plugin format, with skills pointing at the shared skills/ tree — in Cursor Agent chat, install with /add-plugin (or search "securable" in the plugin marketplace, where listed). Independent of the plugin, the always-apply securability kernel rule ships pre-generated at bindings/cursor/securable.mdc; copy it into your project's .cursor/rules/ to bind every generation to the kernel.

Devin

Install from this repository:

devin plugins install Securability-Engineering/securable-claude-plugin

Update later with devin plugins update securable-claude-plugin. The manifest lives at .devin-plugin/plugin.json.

opencode

Tell opencode:

Fetch and follow instructions from https://raw.githubusercontent.com/Securability-Engineering/securable-claude-plugin/refs/heads/main/.opencode/INSTALL.md

Or by hand: .opencode/INSTALL.md — a clone plus scripts/install_skills.sh --target .opencode (project) or --target "$HOME/.config/opencode" (global).

Any AGENTS.md / SKILL.md agent (Codex, Zed, Amp, …)

The skills follow the Agent Skills standard (SKILL.md + YAML frontmatter) and the repo ships a tool-agnostic AGENTS.md entry point, so the pack works in any conforming harness. Tell the agent:

Fetch and follow instructions from https://raw.githubusercontent.com/Securability-Engineering/securable-claude-plugin/refs/heads/main/.agents/INSTALL.md

Or run the installer yourself from a checkout:

# Into the current project (agent-standard path):
scripts/install_skills.sh --target .agents

# opencode project config, or global:
scripts/install_skills.sh --target .opencode
scripts/install_skills.sh --target "$HOME/.config/opencode"

# Claude Code project skills (without installing the plugin):
scripts/install_skills.sh --target .claude

The script copies skills/, data/, plays/, and templates/ (plus schema/, core/, and rules/) together under one root — the layout the skills' internal references depend on.

Developing this plugin

Run claude --plugin-dir . from the repo root so commands and skills load exactly as a plugin install would. Note that the repo's root CLAUDE.md/AGENTS.md is read only when the repo is opened as a project; installed plugin users get the skills and commands, not that file.

Slash Commands

Commands live in commands/ (the plugin-standard location) and are thin dispatchers — each delegates to its skill, which holds the authoritative procedure.

Command	Description
`/securability-review`	Run a full SSEM securability assessment on code
`/secure-generate`	Generate code with FIASSE/SSEM constraints applied
`/prd-securability-enhance`	Enhance PRD features with ASVS + FIASSE/SSEM requirements
`/fiasse-lookup`	Look up FIASSE/SSEM reference material by topic

Example: PRD Enhancement

See the before/after example in:

examples/prd-enhancement/input-prd.md
examples/prd-enhancement/enhanced-prd.md
examples/prd-enhancement/README.md

SSEM Model (FIASSE v1.1)

The Securable Software Engineering Model (SSEM) defines ten attributes across three pillars:

Maintainability	Trustworthiness	Reliability
Analyzability	Confidentiality	Availability
Modifiability	Accountability	Integrity
Testability	Authenticity	Resilience
Observability

Each attribute is scored 0–10 and carries equal weight (1/10 of the overall score). The overall score is the mean of the attribute scores, subject to a weakest-link floor:

raw mean = mean of all assessed attribute scores
floor    = lowest assessed attribute score + 3.0
overall  = min(raw mean, floor)

The floor keeps a single catastrophic attribute from being averaged away — a service that is strong everywhere except input handling is not an adequate service. Pillar scores are reported as diagnostics and never feed the overall score. See skills/securability-engineering-review/SKILL.md for the full rubric, the Not assessed / N/A states, and severity classification.

Scoring conduct: Per FIASSE v1.1 Appendix A.4, a composite SSEM score is a directional management aid for tracking a system against itself over time — not a statement of assurance, compliance, or security. Every report pairs the score with its rationale and a prioritized list of changes.

v1.1 update: "Request Surface Minimization" is now Canonical Parsing (S4.4.1.1) and "Derived Integrity" is now Isolated Integrity (S4.4.1.2). v1.1 also adds the Quality-Security Relationship (S2.4), the Securability Report (S5.2.1–S5.2.5), and scoring guidance (SA.4). Measurement guidance lives in Appendix A (data/fiasse/SA.*.md).

Project Structure

AGENTS.md                          # Canonical agent entry point (AGENTS.md standard, tool-agnostic)
CLAUDE.md                          # Thin stub importing AGENTS.md (project-mode Claude Code)
core/
  kernel.md                        # Securability kernel — source of truth for all bindings
bindings/                          # GENERATED per-harness kernel bindings (never edit)
  cursor/securable.mdc             # Cursor always-apply rule
  copilot/copilot-instructions.md  # GitHub Copilot custom instructions
  gemini/GEMINI.md                 # Gemini CLI context
  aider/CONVENTIONS.md             # Aider conventions
schema/
  securable/                       # JSON Schemas for the securable contract (.securable/*)
rules/
  opengrep/securable.yaml          # Held-check rule pack mapped to the anti-pattern tags
.claude-plugin/
  plugin.json                      # Plugin manifest (Claude Code) — canonical version source
  marketplace.json                 # Marketplace manifest
.cursor-plugin/
  plugin.json                      # Plugin manifest (Cursor) — kept in lockstep
.devin-plugin/
  plugin.json                      # Plugin manifest (Devin) — kept in lockstep
.opencode/
  INSTALL.md                       # Agent-followable install instructions (opencode)
.agents/
  INSTALL.md                       # Agent-followable install instructions (any AGENTS.md/SKILL.md agent)
commands/
  securability-review.md           # /securability-review — thin dispatcher to the review skill
  secure-generate.md               # /secure-generate — thin dispatcher to the generation skill
  prd-securability-enhance.md      # /prd-securability-enhance — thin dispatcher to the PRD skill
  fiasse-lookup.md                 # /fiasse-lookup — thin dispatcher to the lookup skill
.claude/
  settings.json                    # Repo-development permissions (not shipped to plugin installs)
.claudeignore                      # Files excluded from context (repo development only)
.gitignore                         # Excludes test run artifacts (iteration-*/) from VCS
data/
  asvs/                            # OWASP ASVS 5.0 requirement chapters (V1–V17)
  fiasse/                          # FIASSE v1.1 reference sections (S1.x–S8.x + Appendix A as SA.x)
skills/
  securability-engineering/        # Code generation wrapper skill
  securability-engineering-review/ # Code analysis skill
  prd-securability-enhancement/    # PRD securability enhancement skill
  fiasse-lookup/                   # FIASSE/SSEM reference lookup skill
plays/
  code-generation/                 # Step-by-step code generation workflows
  code-analysis/                   # Step-by-step analysis procedures
  requirements-analysis/           # Step-by-step PRD enhancement workflows
templates/
  finding.md                       # Individual finding format
  report.md                        # Full assessment report format
template/
  SKILL.md                         # Template for creating new skills
scripts/
  extract_fiasse_sections.py       # Utility to extract sections from FIASSE v1.1 framework markdown
  install_skills.sh                # Layout-preserving installer for opencode / other agent tools
  build_bindings.py                # Kernel -> bindings generator (--check = CI drift guard)
  validate_securable.py            # Securable-contract validator (shape + semantics + ASVS existence)
  check_refs.py                    # ASVS/FIASSE reference integrity checker
  check_manifests.py               # Per-agent manifest lockstep checker (name/version/license)
  securability_report.sh           # Merge-time Securability Report via any agent CLI
  test_opengrep_rules.sh           # Rule-pack test runner (skips if opengrep absent)
  run_checks.sh                    # Everything CI runs, in one command
  build_plugin_zip.sh              # Release zip builder
  generate_marketplace_json.sh     # Release marketplace manifest builder
examples/
  prd-enhancement/                 # Before/after PRD securability enhancement example
  securable/                       # Worked securable-contract example (requirements + boundaries)
docs/
  critical-review-2026-08.md       # Critical assessment + enhancement plan for this plugin
  securable-contract.md            # The securable contract: files, lifecycle, validation
tests/                             # Regression tests (see Testing below)
  run_tests.py                     # Claude Code CLI test runner (skill workspaces)
  kernel_ab.py                     # Kernel A/B runner + detector self-tests
  README.md                        # Test workspace conventions
  prd-securability-enhancement-workspace/
  securability-engineering-workspace/
  securability-engineering-review-workspace/
  kernel-ab-workspace/             # Kernel A/B evals (naturalistic prompts, deterministic grading)
  securable-contract/              # Contract validator tests
  opengrep-fixtures/               # Paired fail/pass fixtures for the rule pack

Testing

Each skill has a regression workspace under tests/ that exercises it on three realistic prompts and grades the output via LLM-as-judge against an asserted expectation set. Workspaces share a common shape:

tests/<skill>-workspace/
  evals/
    evals.json            # prompts + assertions for each eval
    inputs/               # input fixtures (PRDs / source files)
  skill-snapshot/         # frozen pre-optimization SKILL.md (the baseline)
  iteration-N/            # generated outputs (gitignored)

To run a workspace end-to-end against the live skill plus the snapshot baseline, with grading:

# Requires the `claude` CLI on PATH and Python 3.9+
python tests/run_tests.py tests/securability-engineering-workspace --grade

# Just the live skill, just one eval
python tests/run_tests.py tests/securability-engineering-review-workspace \
    --config with_skill --eval-id 1 --grade

See tests/README.md for full conventions, the per-iteration output layout, and how the runner integrates with the skill-creator aggregator and viewer.

Versioning

The plugin carries its own semantic version, independent of the FIASSE version it targets.

Plugin version (.claude-plugin/plugin.json) — semver for this plugin's own behaviour. A major bump means the review output changed incompatibly. 2.0.0 is the first release of the ten-attribute, equal-weight scoring model with a weakest-link floor. The per-agent manifests (.cursor-plugin/, .devin-plugin/, and plugins[0] in marketplace.json) carry the same version; scripts/check_manifests.py fails CI if they drift, and scripts/build_plugin_zip.sh stamps all of them at release time.
FIASSE target version — the framework release the reference data is extracted from. Recorded machine-readably in the fiasse_version frontmatter of every file under data/fiasse/, and in prose here and in CLAUDE.md.

The two moved together through 1.0.4 and no longer do. Tracking a new FIASSE release is not automatically a major plugin bump, and a scoring change is a major bump whether or not FIASSE moved.

Release tags are v<semver> (e.g. v2.0.0). scripts/build_plugin_zip.sh and scripts/generate_marketplace_json.sh both derive the published version by stripping the leading v; the git ref keeps it.

References

FIASSE Framework v1.1 referenced version
OWASP/FIASSE — Source repository
OWASP/FIASSE — Official Home
OWASP secure-agent-playbook — Larger combined project for agent-oriented guidance
Securability-Engineering — Organization hosting IDE-specific versions of this plugin
loose-notes_claude_aspnet_pva — Worst-case experiment test project

License

CC-BY-4.0 — See LICENSE

Navigation Menu

Uh oh!

Repository files navigation

securable-claude-plugin

Overview

Elevating any harness

Installation

Claude Code

Cursor

Devin

opencode

Any AGENTS.md / SKILL.md agent (Codex, Zed, Amp, …)

Developing this plugin

Slash Commands

Example: PRD Enhancement

SSEM Model (FIASSE v1.1)

Project Structure

Testing

Versioning

References

License

About

Releases

Packages

Used by

Contributors

Languages

Name		Name	Last commit message	Last commit date
.agents		.agents
.claude-plugin		.claude-plugin
.claude		.claude
.cursor-plugin		.cursor-plugin
.devin-plugin		.devin-plugin
.github/workflows		.github/workflows
.opencode		.opencode
.vscode		.vscode
bindings		bindings
commands		commands
core		core
data		data
docs		docs
examples		examples
plays		plays
rules/opengrep		rules/opengrep
schema/securable		schema/securable
scripts		scripts
skills		skills
template		template
templates		templates
tests		tests
.claudeignore		.claudeignore
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md

Uh oh!

Latest commit

History

Folders and files

Repository files navigation

securable-claude-plugin

Overview

Elevating any harness

Installation

Claude Code

Cursor

Devin

opencode

Any AGENTS.md / SKILL.md agent (Codex, Zed, Amp, …)

Developing this plugin

Slash Commands

Example: PRD Enhancement

SSEM Model (FIASSE v1.1)

Project Structure

Testing

Versioning

References

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages