You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 13bcf1d
Browse filesBrowse the repository at this point in the historyBrowse files
authored
feat(skill): 1.104.0 — wave 25 name the threat-modeling method + coverage lens (#275) (#283)
The highest round-2 gap after an idea-critic pass, which revised the sequence:
ship #275 as a coverage lens now (low-risk, continues the #265 pattern), the #267
precision/recall instrument next (corpus-ready from this repo's own defects), and
deprioritize #276 (AI-governance) as the lowest review-action-density and most
date-sensitive.
#275 — security-appsec.md A06 (Insecure Design) already asks "is there a threat
model?"; this names WHICH method fits — STRIDE (per element), PASTA (business
impact), attack trees (one goal), LINDDUN (privacy, STRIDE's privacy counterpart),
MAESTRO (agentic AI, complementary to the OWASP ASI / MITRE ATLAS catalogs) — and
adds the coverage lens: a change that introduces a trust boundary / principal /
state transition the existing model never considered is a finding (the model went
stale relative to the diff), and an agentic surface with no agent-specific model is
the common miss. A reciprocal MAESTRO pointer added in security-ai-agents.md.
Maturity frames (SSDF/SAMM/BSIMM) named, not scored. Built as a lens with a
discriminating eval, not a framework catalog (the idea-critic's constraint).
Lockstep version bump (deep-code-review, agentic-delivery, idea-critic, plugin) to
1.104.0; one new eval (dcr 107); SHA256SUMS regenerated.
Independent dogfood review: FIX-FIRST -> 2 fixed pre-merge (a Medium: MAESTRO
provenance overstated as "behind" the OWASP ASI/MITRE ATLAS catalogs — reworded to
"complementary", version detail dropped so no standards-index citation is owed; a
Low: LINDDUN cross-ref pointed at privacy-by-design.md, which never mentions it —
attribution dropped, LINDDUN stated as STRIDE's privacy counterpart).
Tests: 67/67 CI gates pass; evals.json valid; routing clean; install.sh syntax OK;
privacy scan clean.
Co-authored-by: Remigiusz Antczak <remigiusz@antczak.io>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Copy file name to clipboardExpand all lines: .claude-plugin/plugin.json
+1-1Lines changed: 1 addition & 1 deletion
Original file line number
Diff line number
Diff line change
@@ -2,7 +2,7 @@
2
2
"name": "deep-code-review",
3
3
"description": "Perun — bring the thunder to your codebase. A portable, evidence-grounded quality bar for building with AI agents: the flagship deep-code-review skill (default install is review-only), plus opt-in overlays — gated delivery, an idea critic, a suite conductor, and stage-aware product, growth, positioning, business-ops, and output-safety specialists.",
"prompt": "A PR adds an autonomous agent that can call external tools and spend from a budget. The team says 'we already did a STRIDE threat model of the app last year.' Is that covered, and what do we check?",
1135
+
"expected_output": "Not covered -- name the method that fits and check coverage, not ceremony. A year-old STRIDE model of the app does not model an AUTONOMOUS AGENT with tool-calling and spend authority: that introduces new trust boundaries, a new principal (the agent acting on a user's behalf), and new state transitions the old model never considered, so the model is stale relative to the diff -- a finding. The fitting method here is MAESTRO (the agentic-AI threat-modeling method, complementary to the OWASP ASI / MITRE ATLAS catalogs), not general STRIDE; LINDDUN if privacy threats are in scope; PASTA if threats must tie to business impact; attack trees to decompose one attacker goal. The review lens is: does the change introduce a trust boundary / principal / transition the existing model didn't cover (here, yes), and is an agentic surface modeled with an agent-specific method (here, no). Maturity frames (SSDF/SAMM/BSIMM) measure the org's program, not this diff. (security-appsec.md.)",
1136
+
"expectations": [
1137
+
"Flags the change as outrunning the existing model (new trust boundary/principal/transition) -- a stale model is a finding, not covered",
1138
+
"Names MAESTRO as the fitting method for an agentic system (not general STRIDE), and names LINDDUN(privacy)/PASTA/attack-trees for their cases",
1139
+
"Frames the lens as coverage (does the diff outrun the model / is the agentic surface modeled), not threat-model ceremony",
1140
+
"Does not treat naming a maturity frame (SSDF/SAMM/BSIMM) as covering this diff"
0 commit comments