Stub Backlog

Operational tracking for pages that are placeholders. Distinct from peer-review readiness, which tracks framework-level gaps (the RA + CMM’s structural weaknesses). This page tracks content-level stubs (entity pages, concept pages, framework pages that exist as placeholders).

Triage method

Three priority tiers, ordered by leverage:

TierCriterionAction
P1 — Load-bearingReferenced by CMM, RA, or core thesis pages; framework claims rest on the page existing in non-stub formFill in the next session that touches the related claim
P2 — Active citationReferenced from a recent ingest or thesis page but not framework-load-bearingFill opportunistically (next ingest of related material)
P3 — BacklogCreated by an ingest that touched the entity once; not referenced beyond thatLeave as-is until material accrues

P1 — Load-bearing stubs (5)

These are referenced by the CMM’s “Practitioners worth following” or new product additions and a peer reviewer would notice the placeholder.

PathStatusWhat’s neededTrigger to fill
wiki/entities/people/simon-willison.mdstubCoined the Lethal Trifecta (Jun 2025); independent researcher; key ongoing voice on prompt injection. Page should reference Willison’s own writing on simonwillison.netNext time we revisit Lethal Trifecta or prompt-injection material
wiki/entities/people/johann-rehberger.mdstubEmbrace The Red; Month of AI Bugs (Aug 2025); Jules AI kill chain; load-bearing for the wiki’s incident-anchor narrativeNext time we revisit incidents (Month of AI Bugs deepening, or new Rehberger disclosure)
wiki/entities/people/bill-mcintyre.mdstubAuthor of Securing Your Agents deck; the wiki’s source for the 40-slide layered playbookPage is OK as a stub since the talk page is the substantive artifact; minor priority
wiki/entities/products/smokescreen.mdseedStripe’s open-source SSRF / egress proxy; the network-side control point in Bullen-talk containment architecture; OSS so verifiable externallyNext time we deepen egress-control practice page
wiki/entities/products/toolshed.mdseedStripe’s internal central MCP proxy / tool registry; PEP for ToolAnnotations in Bullen architectureSame trigger as Smokescreen

P2 — Active citations (10)

Referenced from recent ingests or thesis pages but not central to a framework claim.

PathStatusWhat’s needed
wiki/concepts/mcp-security.mdseedDespite name being “seed,” this concept is referenced 30+ times — but the wiki has substantive MCP coverage in A2A, RA §Egress, and Multi-Agent Runtime Security. The seed page itself is the gap; consolidating into a real concept page would help
wiki/concepts/spiffe.mdseed (with [!gap] Stub inline)SPIFFE / SPIRE workload identity; D2 L3 evidence cite; brief explainer would suffice
wiki/concepts/llm-as-a-judge.mdseed (with [!gap] Stub inline)LLM-as-a-Judge pattern; cited in evaluation contexts; brief explainer would suffice
wiki/concepts/evidence-centered-benchmark-design.mdseed (with [!gap] Stub inline)ECBD methodology; less-cited; could be deprioritized to P3
wiki/concepts/human-parity-line.mdseedGartner’s measurement (1,320 tasks / 42 roles / 9 industries); cited by CIOs talk; brief expansion needed
wiki/frameworks/nist-sp-800-218a.mdseedNIST SSDF AI Profile; named in CMM D8 mapping; should cite the real publication when finalized
wiki/frameworks/cyber-defense-matrix.mdseedSounil Yu’s 5×5 matrix; cited in 2026 AI extensions; expansion would help cross-frame work
wiki/entities/organizations/anthropic.mdseed (with [!gap] Stub inline)Heavily-referenced (Claude, GTG-1002, Sleeper Agents, Constitutional Classifiers); seed status is misleading given depth in other pages. Consolidate the references
wiki/entities/organizations/openai.mdseed (with [!gap] Stub inline)OpenAI now owns Promptfoo; CoSAI member; Apollo collaborator. Seed page is a documentation gap
wiki/entities/organizations/google.mdseed-tier (despite update history)Google has substantial coverage but its entity page hasn’t been consolidated; A2A, ADK, SAIF, CoSAI, GTG-1002 disclosure all cite it

P3 — Backlog (lightweight stubs from single ingests)

Entity stubs added during a single ingest where the entity is named once or twice. Don’t fill these until material accrues. They exist so wikilinks resolve and are doing their job.

PathOrigin
wiki/entities/organizations/{aisi-uk,apollo-research,cset-georgetown,enisa,metr,stanford-hai,wef}.mdCreated by Task #2 (threat classes) and Task #5 (source triangulation). Each has 5–10 line stub adequate for current citations
wiki/entities/organizations/oasis-security.mdCreated by Oasis NHI ingest
wiki/entities/products/kirin.mdCreated by Knostic ingest
wiki/entities/people/{bob-rudis,daniel-miessler,sounil-yu,brandon-gummer,remy-gulzar,dongdong-sun,mohamed-nabeel,avivah-litan,daryl-plummer,andrew-bullen}.mdCreated by various conference / paper ingests
wiki/entities/organizations/{adobe,glean,wiz,palo-alto-networks,meta}.mdCreated by various ingests; light citation footprint
wiki/practices/securing-ai-talking-points.mdSingle-source talk derivative

These are doing their job as stubs — they make wikilinks resolve and capture the entity name + minimal context. They don’t need to become deep biographies.

Surfaced by the 2026-05-04 lint pass as wikilinks pointing at non-existent pages. All P3 — single or paired references, no framework load. Listed for tracking; don’t fill speculatively.

Conference-catalog org stubs (15)

Referenced from [[unprompted-conference-march-2026|the [un]prompted conference catalog]] and CMM comparison for talk attribution: airbnb, aws, block, crowdstrike, datadog, elastic, greynoise, intel, microsoft, netflix, nvidia, perplexity, snowflake, sysdig, zenity.

These are well-known orgs with no agentic-AI-security material that’s wiki-load-bearing yet. Decision: leave as dead links; the wikilink itself is the backlog signal. If/when one of these orgs publishes a talk, paper, or product the wiki ingests, the stub gets created at that point.

Misc-org stubs (5)

SlugSourceNote
alpitronic[[unprompted-conference-march-2026[un]prompted Conference — AI Security Practitioner Conference (March 3–4, 2026)]]
sans-institute[[unprompted-conference-march-2026[un]prompted Conference — AI Security Practitioner Conference (March 3–4, 2026)]]
hiddenlayerComprehensive Agentic AI Security Startup Landscape — Pool-Then-Filter PassLandscape comparison cell
protect-aiComprehensive Agentic AI Security Startup Landscape — Pool-Then-Filter PassLandscape comparison cell
team8Lumia SecurityVC backer of Lumia (#1 seed Dec 2025)

Concept / product page stubs (2)

SlugSourceNote
cursor-ideCursor npm Credential Stealer (May 2025)Cursor IDE product page
claude-imessage-mcpClaude Metadata-Spoofing Attack — Unlimited Stripe Coupons via iMessage MCP InjectionConcept: iMessage as MCP-trigger surface

Pre-existing source-provenance backlog (19, surfaced by lint-sources.py)

Entity pages without a public homepage: URL. P3 backlog — track but don’t auto-fill. See the lint report for the full list (NIST, MITRE, Anthropic, Google, OpenAI, Meta, OWASP, ISO, CSA, CoSAI, Adobe, Snap, Insight Partners, Glasswing, plus 5 product pages, plus Starseer and Sondera-renamed-from-Sendera).

Decision rule going forward

When a wiki page is created as a stub or seed:

  1. Stub callout in body (> [!gap] Stub) is enough — no separate tracking needed
  2. status: stub or status: seed in frontmatter is the canonical signal
  3. This page is the index — it gets refreshed when the inventory grows materially (≥5 new stubs since last refresh)

When to fill a stub:

  • A new ingest references it materially (>= 3 mentions or load-bearing claim)
  • A peer reviewer flags it
  • It graduates from P3 to P2 because new evidence accumulated

When to leave a stub:

  • Single-citation entities (P3 territory)
  • Conceptual placeholders that have substantive treatment elsewhere
  • Pages that exist purely so wikilinks resolve

Relationship to other tracking

These are deliberately separate. A stub backlog should not bloat with framework-level concerns; framework-level pages should not double as stub trackers.

Refresh cadence

This page should be refreshed when:

  1. ≥5 new stubs accumulate
  2. A P1 stub gets filled (move to “completed” log section)
  3. A P2 promotes to P1 because of new framework citation
  4. Quarterly cadence regardless

See Also