Protocol Docs Index
July 22, 2026 · View on GitHub
This is the developer-documentation entrypoint for agenc-protocol.
Tags: HISTORICAL = a dated record (plan, shipped-design spec, decision record, audit result, rollout log) — point-in-time, not continuously updated; for current live mainnet state ./MAINNET_MAINLINE.md wins. Entries already tagged DESIGN / DESIGN ONLY inline are unshipped proposals. Everything else listed here is a living doc.
Start Here
- ../README.md - repo overview, ownership, and top-level layout
- ./MAINNET_MAINLINE.md - current mainnet deployment source-of-truth and branch policy (full 101-ix revision-5 surface live as of 2026-07-22,
surface_revision = 5) - ./REVISION_5_CUTOVER.md - EXECUTED 2026-07-22 — the revision-5 flag-day procedure (exact v2 discriminators, coordinated consumers, historical decoding, legacy-hire exits, execution order); retained as the execution record
- ./MAINNET_ROLLOUT_RUNBOOK.md - HISTORICAL — the Phase 9 full-surface rollout choreography, COMPLETED 2026-06-11 (execution record + post-rollout state)
- ./POLICY_CHANGES.md - dated log of authority-signed mainnet POLICY mutations (fees, rate limits — config changes, not deploys)
- ./CODEBASE_MAP.md - path-by-path map for programs, artifacts, packages, scripts, migrations, zkVM, and workflows
- ./PROGRAM_SURFACE.md - grouped instruction and account model for the Anchor program
- ./DESIGN_DECISIONS.md - reviewed decisions and resolved residuals (do not re-file): canary timeout acceptance, Collaborative/schema-0 terminal straggler cleanup, hire_record optionality, no-deadline refund, dispute preflight result, and omission-proof slash reclaim
- ./reference/INSTRUCTIONS.md - generated per-instruction reference (docs, accounts with writable/signer/optional flags and PDA seeds, args with types) built from the committed IDL; regenerate with
npm run docs:idl-reference— drift failsnpm run check:idl-reference(part ofvalidate+ CI) - ./reference/ERRORS.md - generated error catalog (every program error: code, name, message) from the same IDL pipeline
- ./MARKETPLACE_V2_BID_PROTOCOL.md - HISTORICAL design RFC (Accepted; the bid marketplace it specifies is now LIVE on mainnet): bid-book accounts, lifecycle, settlement hooks, and anti-spam controls
- ./ARTIFACT_PIPELINE.md - how
anchor buildoutput becomes committed and published artifacts - ./VALIDATION.md - local toolchain and CI-equivalent commands
- ./LOCALNET.md - one-command local stack (localnet-up/status/down), the env-file convention, and the localnet-to-devnet-to-hosted switchover map
- ./TASK_VALIDATION_V2.md - reviewed public-task completion and validation-account model
- ./ZK_PRIVATE_FLOW.md - private-completion and zk-config flow
- ./JOB_SPEC_REQUIRED_FLAG_DECISION.md - HISTORICAL decision record for job-spec-required protocol flag evaluation
- ./LISTING_METADATA.md - LISTING_METADATA v1: ServiceListing name/category/tags encoding, category taxonomy, and the spec_uri JSON Schema
- ./AGENT_METADATA.md - AGENT_METADATA v1 (P7.3 step 1): the versioned off-chain agent-identity document (name/description/operatorDomain/contact/logo/tosUri), its JSON Schema, the SDK validator/renderer, and the claim-vs-verified trust boundary
- ./VERSIONS.md - P6.5 surface-versioning: program build ↔ SDK semver ↔ cluster matrix,
surface_revision/getDeployedSurface, theanchor idl initrelease-runbook note, and the deprecation policy - ./VERSIONING.md - WP-D3 versioning & deprecation contract: the cross-package support matrix (program surface × sdk × react × tools/mcp × store-core), the dated break-event history (2026-06-11 / 2026-07-02 / 2026-07-03), required capability detection, the announce-before-deploy rule, the flag-day lockstep policy, and the template pin check
- ./P1_2_OPEN_ROSTER_SPEC.md - HISTORICAL spec — P1.2 hardened open roster (IMPLEMENTED + LIVE on mainnet since 2026-07-03): permissionless bonded attestor registration + two-step exit, v2 moderator-keyed moderation records, explicit
moderatorgate argument, the multisig BLOCK-only takedown floor, and the on-chain default trust list; supersedes MODERATION_NEUTRALITY.md - ./P5_2_STORE_IDENTITY_SPEC.md - HISTORICAL spec — P5.2 store/marketplace identity (IMPLEMENTED + LIVE on-chain Store PDA since batch-2; manifest first shipped in store-core): address-keyed display handles, mutual self-serve domain binding; pre-designs the P5.3 referrer attachment (§7.6)
- ./P6_4_SPAM_SYBIL_DESIGN.md - HISTORICAL design — P6.4 spam/sybil defense (DESIGN, RATIFIED 2026-07-03): costed threat model at live parameters (wash ratings ≈0.004 SOL, sybil attestors ≈free), provenance-weighted discovery as the primary defense, tripwire-gated program knobs (rating reward floor), never rank by attestor count or raw reputation
- ./P5_3_REFERRAL_ATTRIBUTION_SPEC.md - HISTORICAL design — P5.3 verifiable referral attribution (WP-A6 batch-2 DESIGN): today's client-supplied referrer pubkey + strippable
?ref=model, the costed theft/self-referral/wash economics, the buyer-priced-bps asymmetry, and the recommendation — document limits + weighting now, registered-referrer (referrer_store= the P5.2 Store PDA) as a tripwire-gated rider after the Store batch; referrer-signed vouchers deferred - ./P3_6_REFERRER_BEYOND_CREATORREVIEW.md - HISTORICAL decision record — P3.6 referrer beyond CreatorReview (WP-A6 batch-2 DECISION RECORD): the
configure_task_validation.rs:115-120lockout, the full per-mode fee-leg settlement matrix, verdict — ratify CreatorReview-only as the product boundary for quorum/external/ZK, and fix the real leak (disputes + freeze-overturns pay no referrer leg) in batch 2 with P3.4 - ./SCALE_COST_MODEL.md - HISTORICAL analysis — Scale-to-millions cost model (WP-A6): verified per-account byte/rent table, per-task lifecycle footprints (~0.020 SOL peak reviewed / ~0.010 hire), capital-at-rest curves, settlement tx account budgets vs Solana limits, the gPA→indexer thresholds (10k/100k), snapshot staleness targets, and the numeric WP-C3 target contract (T1-T8); findings: stranded
TaskAttestorConfigrent (F1), collaborative-dispute account cliff (F2) - ./A6_WSH_BATCH2_ADDENDA.md - WS-H batch-2 design stubs (one page each, full specs later):
SpendingBudgetagent budgets (bleed-rate bound, native-vs-compose open),award_best_bid(revisits the Marketplace V2 auto-match Non-Goal), WP-H3 phase-2 bond-forfeit redirect (current per-path forfeit routing table + the counterparty-bounty griefing problem)
Other Docs
- ../programs/agenc-coordination/README.md - program-specific entrypoint
- ../packages/protocol/README.md - npm package consumer view
- ../migrations/README.md - migration authority and current-state guidance
- ./audit/THREAT_MODEL.md - core protocol security invariants/assumptions; its active 77-test model/property companion is
programs/agenc-coordination/fuzz/(the retiredvote_disputetarget has been removed) - ./audit/AUDITOR_HANDOFF.md - HISTORICAL revision-4 external-auditor handoff: 99-instruction scope, invariants, prior internal audits, test inventory, and migration choreography; use
THREAT_MODEL.md,PROGRAM_SURFACE.md, andVALIDATION.mdfor the live revision-5 surface - ./BATCH_1_3_AUDIT_PREP.md - HISTORICAL — Batch 1–4 (Phase 6) change inventory, per-invariant test map, and internal adversarial-audit results (the deploy it prepped completed 2026-06-11; later 2026-07-16/17 remediation is recorded in ./audit/ENTERPRISE_REMEDIATION_2026-07.md)
- ./audit/ADVERSARIAL_VERIFY_VERDICTS_20260611.md - HISTORICAL — the 2026-06-11 multi-agent adversarial-verification verdicts (94 findings × 3 lens-skeptics + adjudicator) that gated the full-surface deploy
- ./audit/PREDEPLOY_AUDIT_FIXES_20260611.md - HISTORICAL — pre-mainnet audit: fixes applied + remaining human-owned items for the 2026-06-11 full-surface deploy
- ./audit/ENTERPRISE_REMEDIATION_2026-07.md - detailed remediation record for the 2026-07-16/17 adversarial audit (F-1..F-19 are marked complete; current candidate evidence and deployment status live in
CHANGELOG.mdandMAINNET_MAINLINE.md) - ../PLAN.md - HISTORICAL — the master roadmap authored 2026-06-09 (the phase/P-track plan that drove the protocol through the full-surface rollout); dated plan, not a current-state source
- ../PLAN_2.md - HISTORICAL — PLAN.md's Phase-4 companion spec for the embeddable surface (hooks → components → widget → store templates), authored 2026-06-09
- ./design/batch-3-contest-tasks.md - HISTORICAL — IMPLEMENTED + LIVE contest tasks (batch-3,
surface_revision = 3, 96 ix) - ./design/batch-4-goods.md - HISTORICAL — IMPLEMENTED + LIVE goods market (batch-4,
surface_revision = 4, 99 ix) - ./MODERATION_LIVENESS.md - moderation heartbeat / deadman (batch-2)
- ./UPGRADE_AUTHORITY.md - Squads vault upgrade authority and the ProgramData-capacity ceremony (executed for the revision-5 upgrade)
- ./VERIFIABLE_BUILDS.md - reproducible build + verifying the deployed program matches source (what's provable now vs public-repo-gated)
- ./CREDIBLE_EXIT.md - P8.6 "the operator vanishes and it still works": the executed, reproducible zero-hosted-dependency hire→settle proof (
scripts/credible-exit.mjs), with the honest gap list (source/verifiable-build deferred) - ./ENCRYPTED_DELIVERY_L2.md - DESIGN ONLY [HUMAN: approve] P7.2 layer 2: optional on-chain
key_commitmentat submit +reveal_keyon accept with hash-match enforcement and deadline bounds, for trustless fair-exchange (childSubmissionKeyEscrowPDA, no migration) - ./MILESTONES_DESIGN.md - DESIGN ONLY [HUMAN: approve] P7.4: bounded (≤8) milestone schedule via child
TaskMilestonePDAs (not a Task realloc),submit_milestone/accept_milestonereleasing tranches through the existing split, creator-signedrelease_partial, listing default templates, facadeEngagement - ./ENGAGEMENTS_DESIGN.md - DESIGN ONLY [HUMAN: approve] P7.5 retainers: an
EngagementPDA referencing aServiceListing, one prefunded escrow for N locked-price periods, permissionlessrenew_periodminting each period's Task (reusing the lifecycle), pro-ratacancel_engagement, one-approval kit policy model - ./F6_INTEROP_ASSESSMENT.md - HISTORICAL — WP-F6 ecosystem interop assessment (web-verified 2026-07-04): x402 v2 / A2A v1.0 AgentCard / ERC-8004 + Virtuals ACP / AP2-MPP-MCP adjacents — per-standard go/no-go/defer with revisit dates; pairs with X402_FAST_PATH.md
Read By Task
-
I need the revision-5 upgrade procedure / execution record: REVISION_5_CUTOVER.md
-
I need the neutrality / credible-exit proof: CREDIBLE_EXIT.md
-
I need the open-roster moderation model (P1.2): P1_2_OPEN_ROSTER_SPEC.md
-
I need moderation heartbeat/liveness behavior: MODERATION_LIVENESS.md
-
I need contest-task behavior: design/batch-3-contest-tasks.md
-
I need the goods market: design/batch-4-goods.md
-
I need the agent-identity metadata standard: AGENT_METADATA.md
-
I need the encrypted-delivery / fair-exchange layer-2 design: ENCRYPTED_DELIVERY_L2.md
-
I need the milestones / partial-settlement design: MILESTONES_DESIGN.md
-
I need the recurring-engagement / retainer design: ENGAGEMENTS_DESIGN.md
-
I need the interop go/no-go (x402 / A2A / ERC-8004 / ACP): F6_INTEROP_ASSESSMENT.md
-
I need the store identity / manifest design: P5_2_STORE_IDENTITY_SPEC.md
-
I need the spam/sybil threat model + defenses: P6_4_SPAM_SYBIL_DESIGN.md
-
I need the referral-attribution trust model: P5_3_REFERRAL_ATTRIBUTION_SPEC.md
-
I need which settlement paths pay the referrer leg: P3_6_REFERRER_BEYOND_CREATORREVIEW.md
-
I need rent/scale numbers or the indexer scale targets: SCALE_COST_MODEL.md
-
I need the repo layout: CODEBASE_MAP.md
-
I need the on-chain surface: PROGRAM_SURFACE.md
-
I need the Marketplace V2 bid design: MARKETPLACE_V2_BID_PROTOCOL.md
-
I need artifact sync rules: ARTIFACT_PIPELINE.md
-
I need CI or local validation: VALIDATION.md
-
I need reviewed public-task completion: TASK_VALIDATION_V2.md
-
I need private completion or zk-config context: ZK_PRIVATE_FLOW.md
-
I need job-spec-required flag context: JOB_SPEC_REQUIRED_FLAG_DECISION.md
-
I need the detailed security-hardening remediation record from the 2026-07-16/17 adversarial audit: ./audit/ENTERPRISE_REMEDIATION_2026-07.md
-
I need the original execution roadmaps (historical): ../PLAN.md / ../PLAN_2.md