Skip to content

PLAN — P2c A2A Discovery / Card Follow-Up Pack

  • Status: v1 shipped on main (2026-03-25) and now publicly released in v3.4.0 — built-in pack a2a-discovery-card-followup, open mirror, parity tests, and minimal json_path_exists.value_equals engine support merged (§ Implementation order). Phase 0 decisions below remain the contract for this pack; future rule changes require PLAN updates.
  • Date: 2026-03-25
  • Owner: Evidence / Product
  • Prerequisite: G4-A Phase 1 — payload.discovery on emitted canonical A2A evidence — merged on main.

Sequencing discipline (why this order)

This wave follows the same discipline that made G4-A strong:

  1. Evidence seam (adapter emits bounded shapes) — done (payload.discovery).
  2. Freeze (normative semantics) — done (G4-A-PHASE1-FREEZE.md).
  3. Pack productization (this PLAN) — plan and freeze first, then YAML + tests in a later PR.

Phase 0 is complete — see § Phase 0 — decisions (locked). Do not commit pack YAML or rules until the implementation PR follows the sequencing in § Implementation order.

Relation to G4-A payload.discovery

G4-A (evidence) P2c (pack)
Defines what appears on /data/discovery/* and bounded meaning (§2b, §4). Defines which lint rules evaluate those fields in bundles — no new semantics and no stronger claims than the freeze.
Adapter-first, Assay-namespaced attributes.assay_g4. Reads canonical emitted JSON only; does not invent upstream protocol types.

Normative semantics for each field remain in G4-A; P2c help_markdown and pack disclaimer must not contradict §2b.

Relation to P2b (companion pack)

  • P2b asserts canonical A2A event-type presence (event_type_exists on assay.adapter.a2a.*).
  • P2c reads G4-A payload.discovery on those same events.
  • The packs may be used together; they are semantically distinct (different product surfaces — not duplicate claims on the same signal).

Goal

Ship a small, bounded companion pack on assay.adapter.a2a.* evidence that can observe G4-A payload.discovery signals — parallel in spirit to P2b, not an “A2A security”, trust, or verification pack.

In scope (v1)

  • Rules that are evidence-only and visibility / presence aligned to G4-A (see § Rule IDs — freeze target).
  • Composition with P2b (a2a-signal-followup): P2c must not duplicate P2b’s three event_type_exists rules unless product explicitly wants redundant checks.

Out of scope

Topic Reason
Card authenticity, signature validity, trusted provenance Not in G4-A v1 seam; pack must not imply them.
Auth / authz correctness, G3-style claims on A2A No G3-shaped evidence on A2A adapter output.
Broad “A2A security” or protocol marketing claims Same bar as P2b non-goals.
Pack engine / ENGINE_VERSION bump Only if Phase 0 proves existing check types are insufficient — default is no bump.
New adapter or evidence signals in the P2c wave P2c is pack-only after G4-A; adapter changes belong to a different wave.

Pack identity — freeze target

Field Freeze value
Pack name (YAML) a2a-discovery-card-followup
Built-in id Same string as name (match P2b pattern)
Version 1.0.0 at first ship (semver; bump only on rule/semver-worthy change)
Kind security (same family as P2b). Product confirms this categorization remains appropriate for observed visibility-only rules (no alternate taxonomy debate in this plan). At implementation, YAML / help_markdown / disclaimer must not imply generic “A2A security” outcomes solely because kind is security — keep outward copy as narrow as the rules.
Rule count (v1) 2 — A2A-DC-001 and A2A-DC-002 only; DC-003 is not v1.

Rule IDs — freeze target (2 rules)

Prefix A2A-DC (D**iscovery **C**ard follow-up) avoids collision with P2b **A2A-001..003.

ID Intent (one line) Bounded meaning (may imply) Must not imply
A2A-DC-001 Agent card visibility observed — at least one canonical A2A event has agent_card_visible: true on payload.discovery (per G4-A paths). Producer asserted the allowlisted visibility flag as true with valid upstream shape (see G4-A §4.1). Card authentic, unspoofed, complete, or “real”.
A2A-DC-002 Extended card access visibility observed — at least one canonical A2A event has extended_card_access_visible: true on payload.discovery (per G4-A paths). Producer asserted the allowlisted extended-card-access visibility flag as true with valid upstream shape (see G4-A §4.2 / §2b). Auth succeeded, authz held, access legitimate, client trusted.

Deferred / opt-in only: A2A-DC-003

Not part of P2c v1 freeze or default ship. Consider only if product explicitly opts in and this document is updated.

Aspect Content
Intent (sketch) Source-kind visibility discipline — e.g. require agent_card_source_kind not unknown when product wants “kind resolved”, or another narrow predicate.
Bounded meaning Which precedence class is recorded (per G4-A §3).
Must not imply Correctness or trustworthiness of that source.

Opt-in gates (all required):

  1. Explicit product justification (user-visible value for “source-kind resolved” or equivalent).
  2. Phase 0 shows the rule is achievable without judgment / quality semantics.
  3. Wording stays strictly non-judgmental (no implicit “good” vs “bad” source).

Explicit non-rules (v1)

These are not P2c v1 rules (list may extend at freeze):

  • Any rule that treats discovery as verification or security outcome.
  • Any rule requiring signature_material_visible: true (v1 adapter always emits false per G4-A §1). P2c v1 does not reopen G4-A’s deferral on signature_material_visible semantics — Phase 0 must not treat signature visibility as an open design question.
  • Duplicates of A2A-001..003 (P2b) unless explicitly desired.

Phase 0 — Pack engine discovery

Phase 0 was a discovery pass on pack-engine fit, CI semantics, and release floors. Decisions are locked in § Phase 0 — decisions (locked). The table below records the original questions; the Done when column reflects the locked outcome.

# Question Done when
1 Can pack checks express boolean JSON true on G4-A-frozen /data/discovery/... paths for DC-001 / DC-002 — i.e. value true, not merely “path/key exists”? SPEC today: json_path_exists is presence-only (a false value still satisfies “path exists”), and conditional is an if/then required-path pattern — it does not mean “at least one event satisfies == true” on its own. Locked intent: implementation PR adds a minimal engine affordance for value equality at a frozen pointer (e.g. optional value_equals on json_path_exists) or an equivalent narrow check; no broad ENGINE_VERSION bump unless unavoidable. Locked: Prefer smallest change that makes == true honest; no gratuitous engine bump (see decision 2 under § Phase 0 — decisions (locked)).
2 Pre-G4-A bundles (no /data/discovery): default fail in normal lint for rules that require discovery, consistent with chosen check types — unless the team explicitly chooses skip/N/A in Phase 0 and documents why (CI consistency). Disclaimer + help_markdown must state the pack requires G4-A discovery emission. Default is for CI clarity and must read consistently across rule descriptions, help text, and example outputs (avoid plan-says-fail vs help-reads-N/A). This product/CI default means a rule reports missing required discovery evidence for this pack, not that the bundle is generally malformed or invalid. Locked: Fail by default for DC-001 / DC-002; disclaimer + help explicit (see decision 3 under § Phase 0 — decisions (locked)).
3 requires.assay_min_version — do not assume the same floor as P2b (e.g. >=3.2.3) by default. Anchor to the first Assay release that ships G4-A payload.discovery in release binaries/artifacts. The exact semver string stays intentionally unset in this plan until pack ship; set in YAML + release notes at ship time. Meaning of requires floors: MIGRATION-TRUST-COMPILER-3.2.md. Locked: String unset in this plan until ship; floor = first G4-A-capable release, not P2b’s floor (see decision 4 under § Phase 0 — decisions (locked)).
4 evidence_schema_version Locked: 1.0, aligned with other built-ins, unless MIGRATION-TRUST-COMPILER-3.2.md or SSOT gives a reason to change (see decision 5 under § Phase 0 — decisions (locked)).

Phase 0 — decisions (locked)

Locked: 2026-03-25. These decisions are product/engine contract for the P2c implementation PR; they do not change G4-A normative semantics.

One-line summary: Ship a minimal two-rule P2c on the G4-A discovery seam (DC-001, DC-002); minimal engine delta so == true is honest (see decision 2); fail pre-G4-A bundles by default for CI clarity; keep assay_min_version unset in this plan until the first G4-A-capable Assay release at pack ship; evidence_schema_version = 1.0; defer DC-003 and any signature/provenance semantics beyond G4-A.

  1. v1 stays exactly two rules. Ship only A2A-DC-001 (agent_card_visible == true) and A2A-DC-002 (extended_card_access_visible == true). A2A-DC-003 remains deferred / opt-in. Rationale: A2A positions Agent Cards and extended card retrieval as core discovery surfaces; recent agent-security work (e.g. A2ASecBench-style emphasis on discovery/card spoofing as a risk surface) supports visibility-on-the-seam first, not an extra “source-kind discipline” rule that drifts toward judgment semantics in v1.

  2. No gratuitous engine bump; smallest change for honest == true. Raw json_path_exists (SPEC) does not distinguish true vs false — only path presence. Locked: the P2c implementation PR includes a minimal pack-engine extension (or equivalent) so rules evaluate JSON boolean true at the frozen /data/discovery/... pointers, without misleading “key present” semantics. Rationale: reuse narrow primitives; avoid a new policy DSL — but do not pretend the pre-P2c engine already enforced value equality.

  3. Pre-G4-A bundles: fail by default in normal lint. Bundles without /data/discovery cause DC-001 / DC-002 to fail (consistent with chosen checks). Disclaimer and help_markdown must state the pack requires G4-A discovery emission. Skip/N/A only if explicitly chosen and documented with rationale. Rationale: clearest CI signal (“this pack needs newer evidence”), not a general claim that the bundle is invalid.

  4. assay_min_version: no exact version string in this plan. At ship, set the floor to the first Assay release that includes G4-A payload.discovery in release binaries/artifacts — not P2b’s substrate floor. Rationale: P2c depends on the G4-A seam, so the floor must track that seam.

  5. evidence_schema_version: 1.0, aligned with other built-ins, unless migration / SSOT gives an explicit reason to differ.

Non-normative context (March 2026): Industry and research emphasis on identity, authorization, auditing, and observed evidence for agents aligns with small, observable pack rules — not broad “secure agent” marketing claims. This plan stays within § Explicit non-rules and does not reopen G4-A’s deferral on signature_material_visible.

Acceptance — plan freeze vs implementation

Gate Criteria
Plan / freeze acceptable Pack identity table filled; 2 rule IDs (DC-001, DC-002) with bounded meaning rows; Deferred DC-003 documented as not v1; Phase 0 decisions locked; non-rules listed.
Implementation PR ready Built-in YAML + packs/open/ mirror byte-for-byte parity (P2b pattern); assay-evidence tests for open/builtin equivalence; requires match release truth; docs (ROADMAP / this PLAN) synced.

Parity, release floor, and docs sync

Implementation order (implementation PR)

Phase 0 is locked (§ Phase 0 — decisions (locked)). Steps 1–4 below are complete for P2c v1 on main (merge 2026-03-25).

  1. Frozen Phase 0 answers + final rule table — done (this PLAN on main).
  2. YAML (built-in + open mirror) — done (crates/assay-evidence/packs/a2a-discovery-card-followup.yaml, packs/open/a2a-discovery-card-followup/).
  3. Tests (parity + bundle fixtures with payload.discovery) — done (crates/assay-evidence/tests/a2a_discovery_card_followup_pack.rs and related).
  4. Docs sync and release note — done (this PLAN + CHANGELOG [Unreleased]; ROADMAP checklist).

Further changes to rules or semantics require updating this PLAN (no ad-hoc pack edits without plan alignment).

References

Changelog

Date Change
2026-03-25 Skeleton: file PLAN-P2c-A2A-DISCOVERY-CARD-FOLLOWUP-PACK.md; sequencing (evidence→freeze→pack); scope/non-goals; pack identity; A2A-DC-001..003 with bounded-meaning rows; Phase 0; acceptance/parity/release floor; implementation order.
2026-03-23 Review pass: v1 = 2 rules in main table (DC-001, DC-002); DC-003 moved to Deferred / opt-in; Phase 0 — boolean true on frozen paths, pre-G4-A fail default + CI copy consistency, assay_min_version anchored to G4-A seam (unset until ship); Relation to P2b; kind: security one-line product note; implementation order clarified (no YAML before freeze/Phase 0; remove “YAML last” ambiguity).
2026-03-23 Polish: implementation guard — step 2 (YAML) blocked until review + Phase 0; Phase 0 row 2 — fail = missing pack-required discovery, not general bundle invalidity; DC-002 bounded meaning parallel to DC-001; kind: security outward-copy caution; explicit non-rule — P2c v1 does not reopen G4-A signature_material_visible deferral.
2026-03-25 Phase 0 locked (+ Copilot #947 follow-up): v1 = DC-001 + DC-002 only; DC-003 deferred; fail pre-G4-A bundles by default; assay_min_version unset until ship; evidence_schema_version = 1.0; status → Phase 0 locked; implementation order → step 2 next. Clarify row 1 / decision 2: SPEC json_path_exists is presence-only; P2c implementation PR needs minimal value-equality (or equivalent) for honest == true — not “existing checks already do this.”
2026-03-25 Implementation sync: pack ships requires.assay_min_version: ">=3.3.0" (built-in + open); SPEC value_equals — JSON equality only (no coercion); parity section updated.
2026-03-25 Shipped on main: P2c v1 implementation merged — built-in a2a-discovery-card-followup (A2A-DC-001 / A2A-DC-002), json_path_exists optional value_equals, ROADMAP / CHANGELOG / this PLAN post-merge sync.