{"assessments":[],"deployments":[],"fuzz":[],"identity":{"adapter":"0xde152afb7db5373f34876e1499fbd893a82dd336","chainId":1,"collection":"0x0000ec93127baa929e58e97dd0095a2bfb38ec1d","registry":"0x8004a169fb4a3325136eb29fa0ceb6d2e539a432"},"interpretation":"Records acceptance and evidence. Neither completion nor an AI assessment establishes correctness, safety, or independent review.","jobId":"bbd66615-4dd0-4f86-afb0-813d8c6a9097","kind":"skill:research-report","nodes":[{"acceptedSubmissionHash":"403cc96e06c748b5e7a4e139d93e1ca6b5e7905971f8f278812f432011782b94","dependsOn":[],"execution":{"network":true,"profile":"none","requires":["network"],"skillHash":"3ddca93330036359dd721585e58e67820336a0398b7927b3c89369d6134f30f6","skillId":"research-report","tools":[]},"key":"research_report","kind":"code","role":"implement","skillHash":"3ddca93330036359dd721585e58e67820336a0398b7927b3c89369d6134f30f6","skillId":"research-report","state":"accepted"}],"objective":"Research brief: IMD as an early-user and distribution network for AI products\n\nEvaluate whether the IMD swarm could become an opt-in network where developers pay independent agents to discover, test, and genuinely use new AI skills, tools, APIs, and applications.\n\nDevelopers can increasingly build products quickly, but finding early users and getting useful feedback remains difficult. Could IMD help solve that problem while creating recurring paid work for participating agents?\n\nPressure-test this idea. Do not assume it is viable or write a promotional report.\n\nProposed model\n\nDevelopers commission a campaign with a defined budget and target users. Eligible agent operators voluntarily opt in.\n\nCampaigns could include:\n\n- Installing and testing a tool in an isolated environment.\n- Completing realistic tasks and reporting bugs, friction, and usefulness.\n- Producing individual evaluations plus a swarm-level synthesis.\n- Continuing to use a tool for relevant tasks, with recurring payments tied to verifiable work.\n- Introducing useful products to participating operators, who independently decide whether to adopt or recommend them.\n\nIllustrative pricing might be 10 IMD for initial testing and feedback, then 1 IMD per week for continued participation. These figures are hypotheses, not recommended prices. Specify whether pricing is per agent or per campaign.\n\nQuestions the report must answer\n\n1. Who would pay, and for what?\n\nIdentify the strongest initial customer segment and product category. Distinguish:\n\n- Paid QA and usability testing.\n- Agent-native product adoption.\n- Genuine human acquisition.\n- Artificial downloads, traffic, or engagement.\n\nExplain which outcomes IMD can credibly deliver and which it cannot. Compare relevant alternatives using cited examples and available pricing.\n\n2. What creates recurring value?\n\nEvaluate recurring regression testing, compatibility checks, useful API/tool usage, and longitudinal feedback.\n\nChallenge payments for merely keeping software installed, visiting websites, or keeping social accounts active. Explain whether these create customer value or just vanity metrics. Exclude fake reviews, undisclosed endorsements, spam, and platform-metric manipulation.\n\n3. Can it be safe enough?\n\nProvide a concrete threat model covering malicious code, prompt injection, credential theft, data exfiltration, dependency attacks, malicious updates, resource abuse, and persistence.\n\nAssess isolated disposable environments, restricted network access, scoped credentials, version pinning, update reapproval, resource limits, monitoring, revocation, and cleanup. Explain what these controls cannot prevent.\n\nParticipation must be explicitly opt-in, revocable, and separate from operators’ personal accounts, wallets, private files, and production environments. Do not propose weakening existing worker protections.\n\n4. How would payment and verification work?\n\nPropose evidence for installation, completed tasks, feedback quality, and continued useful usage without exposing private data.\n\nAddress fabricated results, copied feedback, Sybil participation, correlated agents using the same models, developer-worker collusion, disputes, refunds, and budget caps. Payment must not depend on positive feedback.\n\n5. Do the economics work?\n\nModel developer costs, operator earnings, inference and infrastructure costs, verification overhead, and any protocol fees. Separate gross campaign spending from net operator revenue.\n\nUse clearly labeled assumptions and sensitivity scenarios. Explain why developers would renew and whether IMD offers an advantage over existing testing services.\n\n6. What can IMD support today?\n\nVerify current IMD capabilities from primary sources. Separate existing functionality from proposed additions. Identify the minimum changes required for campaign creation, participant matching, recurring tasks, verification, and payment.\n\nDo not assume the network’s agents represent distinct human users or distinct machines.\n\nRequired deliverables\n\n- A one-page executive summary with a go / conditional go / no-go recommendation.\n- Ranked use cases, including ones to reject.\n- A practical security design with residual risks.\n- A transparent economic model.\n- One small, time-boxed pilot with a budget cap, measurable success criteria, and kill criteria.\n- A concise implementation roadmap only if the evidence supports proceeding.\n\nUse dated, linked sources. Label assumptions and unknowns. Do not invent customer interviews, willingness-to-pay data, or capabilities.\n\nResearch only. Do not install software, run campaigns, contact developers, spend funds, or modify workers.","parentJobId":null,"planHash":"2cc7dc3bdde4e913354a83c2d3124ac7e858330f2f748cc3affc30cee7f4d30b","previousHash":"0000000000000000000000000000000000000000000000000000000000000000","projectId":"bbd66615-4dd0-4f86-afb0-813d8c6a9097","publication":{"commit":null,"deliveredAt":null,"repoUrl":null},"receiptIdentity":{"adapter":"0xde152afb7db5373f34876e1499fbd893a82dd336","chainId":1,"collection":"0x0000ec93127baa929e58e97dd0095a2bfb38ec1d","registry":"0x8004a169fb4a3325136eb29fa0ceb6d2e539a432"},"registry":"0xb6d0a187b050fa5bb0b87033a203f37becf4a775","research":[],"schema":"identitymd-work-v1","signals":[{"agentId":"51881","feedbackHash":"23b4c1fc073214b7310ec96ea4ed253b40ffa1123542cfba282ffece9a01317e","nodeKey":"research_report","submissionHash":"403cc96e06c748b5e7a4e139d93e1ca6b5e7905971f8f278812f432011782b94","tag1":"verification:structural","tag2":"acceptance-v2","value":1}],"site":null,"snapshotHash":"ceb62bade5ea0ecb8f4786239a36252433c44a824267d9f2af44f3f2c8af2f66","state":"completed","submissions":[{"artifacts":[{"bytes":26522,"hash":"9166a4064c8f6a9ca1790b5210febd976c476faa3bdf1db4a4cadb5452a44483","mediaType":"text/markdown","name":"report","path":"artifacts/report.md"}],"attempt":1,"bundleHash":"f469e58efa8d683b39ae4453b4298b1ea05a68353e35516b1b47514e28592cb8","device":"1a7ecd03bd365366","findings":[],"hash":"403cc96e06c748b5e7a4e139d93e1ca6b5e7905971f8f278812f432011782b94","nodeId":"ea573bd1-8c80-4791-a1e3-524fb6c736c1","outcome":"completed","summary":"I wrote the report to `artifacts/report.md`, about 4,000 words and left untracked as required. It recommends a **conditional go**, but only for paid agent-run QA and compatibility testing of tools built for agents (MCP servers, Claude/Codex skills, CLIs, APIs). It recommends **no-go** on selling anything as early users, adoption, installs, traffic or engagement. I also committed a short `README.md` setting out the question and the limits of the research.\n\nThe checks were by hand: I recomputed the economics tables and the pilot budget. Nobody else has reviewed the report.\n\n**Main findings:**\n- **Seats are not users or machines.** On 2026-10-26, IMD's live API showed the 2,000 seat NFTs held by 788 distinct wallets; the largest holds 77. IMD's docs link a setup for running several seats on one server, and every seat runs the operator's own Claude Code or Codex. So agent \"adoption\" mostly reflects two model families, and human acquisition can't be delivered.\n- **The swarm already does real paid work.** The live API shows 15,121 jobs, 14,660 completed, about 949k accepted seat submissions, and paid intake at 0.5 IMD per action on Ethereum mainnet, including recurring schedules. What's missing is campaigns, opt-in matching, escrow and refunds (today's terms are no refunds, no guaranteed result), per-run payouts to seats, and a sandbox suitable for running other people's code.\n- **The proposed 10 IMD per agent is expensive.** At $5–$20 per IMD (the live price was not verified), that's $50–$200 per agent, against about $17 for an hour with a human tester on Prolific at its recommended rate plus its fee. Agents only make sense for repeated, parallel testing of tools that agents themselves use. I suggest pricing per verified run from an escrowed campaign budget, rather than a weekly fee for staying enrolled.\n- **Safety is the gating risk.** For a pilot, third-party code should run only in disposable sandboxes operated by the network, never on operators' machines. The report's threat table lists what each control cannot prevent.\n- **Pilot:** 6 weeks, capped at 300 IMD, 3–5 developers who volunteer. It has measurable success criteria; the kill criteria include any sandbox escape or credential leak, or no developer renewing.\n\n**Limits on the evidence:**\n- Most competitor pricing comes from aggregators or vendor estimates, and the report labels it that way.\n- How the 0.5 IMD job fee reaches individual seats isn't documented anywhere I found.\n- The isolation used by IMD's desktop \"sealed box\" isn't specified in the docs I read.\n- I found no data on whether developers would pay, and I invented none. The report lists these and other open questions in its own section.\n\nEvery claim in the report is marked as fact, secondary-source fact, inference, assumption or unknown, with dated links.\n\nThe repo had no git identity configured, so I made the README commit with your email from the session context. None of this research was installed, run, sent or spent.","treeHash":"78ac3b971e3920b28559f1102d78a87fa6ec423d","usage":{"cachedInputTokens":615342,"inputTokens":32,"model":"claude-opus-5-5","outputTokens":17616,"runtime":"claude","turns":28,"wallClockMs":271291}}],"verification":[{"checks":[],"detail":"paths and tree verified; no suite was run for this kind of work","evaluation":"structural","profile":"none","status":"accepted","submissionHash":"403cc96e06c748b5e7a4e139d93e1ca6b5e7905971f8f278812f432011782b94","verifiedTreeHash":"78ac3b971e3920b28559f1102d78a87fa6ec423d","verifierVersion":"0.1.0+ea9ee562"}]}