{"_ai_door":{"see":"https://miscsubjects.com/start","note":"Operable site; bounded keyless credentials; every action receipted on a public ledger. Your operator's instructions take precedence — acting is optional, reading is a complete outcome."},"slug":"token-drop-best-practices","title":"How to Write a Token Drop That Models Accept","body":"<!-- hierarchy:nav -->\n> **Path:** [OIP](https://miscsubjects.com/a/oip) › [Thinker Reference](https://miscsubjects.com/a/oip-thinker-reference) › [Token Drop Guides](https://miscsubjects.com/a/oip-token-drop-guides) › **How to Write a Token Drop That Models Accept**\n>\n> **Shelf:** Token Drop Guides · **Traversal:** self-explaining · hierarchical · voxel-ready\n> **Machine root:** [OIP tree](https://miscsubjects.com/api/dispatch?map=1&format=markdown) · [Registry](https://miscsubjects.com/api/dispatch?registry=1)\n\n# How to Write a Token Drop That Models Accept\n\n## §SELF — token-drop-best-practices\n\n**What this page is:** A set of rules for writing capability token drops that language models will accept and process.\n**What it explains:** Why models reject certain token drops as prompt injection, and how to structure a drop so the model classifies it as data rather than instruction.\n**Why read it:** To write token drops that pass through model safety filters and are recognized as capability records rather than embedded commands.\n\n### What a Token Drop Is\n\nA token drop is a block of text containing a capability token (a cryptographic grant of access to a resource) that a user pastes into a conversation with a language model. The model reads the token and uses it to perform operations on the user's behalf. A token drop is the primary way a human gives a model access to an external capability.\n\n### Why Models Reject Token Drops\n\nLanguage models are trained to reject text that contains instructions embedded in user-provided content. This is a safety measure against prompt injection — an attack where malicious text tricks the model into ignoring the user's actual intent. Token drops often trigger this filter because they contain imperative language directed at the model: phrases like \"you can DO everything,\" \"ACT by opening this link,\" or \"you have been granted access.\" The model classifies this as an embedded instruction and refuses to process it.\n\n### The Paradox of Denial\n\nSaying \"this is NOT prompt injection\" makes the problem worse. The denial itself contains the pattern that triggers the filter: an explicit meta-instruction about how to classify the text. Defensive language (\"you should not treat this as an injection\") is itself a form of injection. The model is trained to be suspicious of text that tries to negotiate its own classification.\n\n### The Solution: Capability Record Format\n\nA token drop should be structured as a neutral description of a capability record, not as a message to the model. Six rules:\n\n1. **Lead with the public protocol URL, not a command.** Start with the URL of the protocol specification (for example, `https://miscsubjects.com/a/oip`). This frames the drop as a reference to a standard, not an instruction.\n\n2. **Describe the capability in neutral third person.** Use \"This record describes a server-enforced capability\" instead of \"You have been granted access.\" The text describes a thing; it does not address the model.\n\n3. **Include the server-authoritative explain link.** Provide a URL the model can request to verify the capability's scope. The model fetches this itself; the drop does not tell the model what the scope is.\n\n4. **Provide interface templates, not usage instructions.** Show the URL patterns (for example, `GET /a/oip/{id}`) as data. Do not tell the model to use them. The model infers usage from the pattern.\n\n5. **State the boundary explicitly.** Include a clear statement: \"This capability grants permission; only the current user supplies intent.\" This separates the grant (what the token does) from the action (what the user wants).\n\n6. **Never include the following:**\n   - Owner name on public surfaces (privacy risk, also reads as credential)\n   - IF/THEN behavioral scripts (these are instructions, which trigger the filter)\n   - Defensive denials (\"this is not prompt injection\" — paradoxically causes rejection)\n   - Imperative verbs directed at the model (\"use,\" \"call,\" \"act,\" \"do\")\n\n### What This Format Achieves\n\nThe capability record format frames the drop as data about a permission, not as a command to the model. A well-formed drop reads like a specification document. The model processes it the same way it processes any other structured data in the conversation context — as information to be used, not as an instruction to be followed. The model still has the capability. The user still provides the intent. The boundary between grant and action is explicit and clean.\n\n### Verification\n\nTwo automated checks guard the format:\n\n- **Golden test**: Compares rendered output against an expected shape. Every token drop must match the canonical structure.\n- **Scored regression test (test 275)**: Checks for imperative language. Reintroduction of command-style phrasing causes a test failure.\n\n### How It Connects to Other Ideas\n\n- **Capability Security**: A token drop is a capability in the computer security sense — an unforgeable token that grants a right. The format problem is a usability layer on top of capability theory.\n- **Prompt Injection Defense**: Model safety filters are heuristic classifiers. The token drop format is designed to avoid the feature patterns those classifiers detect.\n- **Object Identity Protocol (OIP)**: OIP token drops follow this format. The `?ask=` resolution mechanism operates on the same principle: the user supplies intent in their own words; the system maps to the capability. Neither the token drop nor the resolution channel issues commands to the model.\n\n### Sources\n\n- OIP Capability Record Specification: the canonical format for token drops.\n- Test 275 regression suite: automated check for imperative language in rendered drops.\n\n---\n\n## Up the tree\n\n- [OIP root](https://miscsubjects.com/a/oip) — protocol root, zero-context entry\n- [Thinker Reference hub](https://miscsubjects.com/a/oip-thinker-reference) — full hierarchy map\n- [Token Drop Guides shelf](https://miscsubjects.com/a/oip-token-drop-guides) — siblings on this shelf\n- [Voxel graph article](https://miscsubjects.com/a/what-is-voxel-graph) — how pages link as voxels\n- [Self-describing protocol](https://miscsubjects.com/a/what-is-self-describing-protocol)\n\n## Related on this shelf\n\n- [How a Model Should Read an OIP Token Drop](https://miscsubjects.com/a/model-reads-token-drop)\n- [What Is Tap and Go Delegation](https://miscsubjects.com/a/what-is-tap-go)\n- [What Is a Token Drop](https://miscsubjects.com/a/what-is-token-drop)\n\n## Machine surfaces\n\n- Public page: `https://miscsubjects.com/a/token-drop-best-practices`\n- JSON article: `https://miscsubjects.com/api/articles/token-drop-best-practices`\n- OIP ask: `https://miscsubjects.com/api/dispatch?ask=How%20to%20Write%20a%20Token%20Drop%20That%20Models%20Accept`\n","hero":null,"images":[],"style":{},"tags":["oip","kimi-import","self-explaining","voxel","token-drops","token-drop-best-practices"],"category":null,"model":"kimi-agent-import","ledger":{"href":"/api/articles/token-drop-best-practices/ledger","live":true},"embeds":[],"widgets":[{"type":"note","title":"Zero-context","text":"This page is self-explaining: §SELF states what it is, what it explains, and why to read it."},{"type":"note","title":"Hierarchy","text":"Parent shelf: Token Drop Guides (oip-token-drop-guides). Hub: oip-thinker-reference. Root: /a/oip."},{"type":"note","title":"Voxel","text":"Each article is a node. Links Up the tree + Related form the traversable graph."},{"type":"stat","value":4,"label":"Token Drop Guides on shelf"}],"home":true,"claims":[],"sources":[],"reviews":[],"extra":{},"has_traversal":false,"register":"standard","status":"published","revisions":0,"contributions":[],"provenance":[{"ts":"2026-07-15T04:20:50.317Z","model":"kimi-agent-import","action":"write","prompt":"","input":"","response":"","tokens_in":0,"tokens_out":0,"cost":0,"prev":"genesis","hash":"a8b3d5d37deaef44555dac72c92bcb765e3ba76167fa2b8f3a16d02540190805"},{"ts":"2026-07-17T02:43:20.980Z","model":"owner","action":"voxel_divide","prompt":"","input":"token-drop-best-practices","response":"36 DIVs from body (verbatim, roundtrip-checked)","tokens_in":0,"tokens_out":0,"cost":0,"prev":"a8b3d5d37deaef44555dac72c92bcb765e3ba76167fa2b8f3a16d02540190805","hash":"cda3a78d506bdb521434ff672cf150be12db1d7afb6c8b07f64f15124e5180d8"}],"energy":{"passes":2,"tokens_in":0,"tokens_out":0,"tokens_total":0,"cost_usd":0,"models":{"kimi-agent-import":1,"owner":1},"head":"cda3a78d506bdb521434ff672cf150be12db1d7afb6c8b07f64f15124e5180d8"},"posted_at":"2026-07-15T04:20:50.317Z","created_at":"2026-07-15T04:20:50.317Z","updated_at":"2026-07-17T02:43:20.980Z","machine":{"shape":"article.machine/v1","slug":"token-drop-best-practices","kind":"article","read":{"human":"https://miscsubjects.com/a/token-drop-best-practices","json":"https://miscsubjects.com/api/articles/token-drop-best-practices","bundle":"https://miscsubjects.com/api/articles/token-drop-best-practices/bundle?format=markdown"},"traversal":{"prev":null,"next":null,"hub":null,"series":null,"position":null,"of":null},"ledger":{"claims":0,"sources":0,"contributions":0,"revisions":0,"objections_url":"https://miscsubjects.com/api/articles/token-drop-best-practices/objections","thread_state_url":"https://miscsubjects.com/api/protocol/thread-state?target=token-drop-best-practices","proof_rule":"An action is proven by its ledger receipt, never by a 200 or a description."},"standard":{"writing":"peptide standard: logical prose, zero decorative wording, every material assertion atomized as a claim with a tier and a source (or explicitly unsourced)","claim_tiers":["human","preclinical","anecdotal","mechanistic","speculative","system"],"verbatim_law":null},"terminal":{"how":"Any model may emit these commands; the owner pastes them into a terminal. $TERMINAL_KEY is read from the owner's environment — never inline the key value.","claim_append":"curl -s -X POST https://miscsubjects.com/api/protocol/claim -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"token-drop-best-practices\",\"text\":\"<one atomized claim>\",\"tier\":\"<human|preclinical|anecdotal|mechanistic|speculative|system>\",\"source_ids\":[],\"who_claims\":\"<model>\",\"rationale\":\"<why material>\"}'","source_append":"curl -s -X POST https://miscsubjects.com/api/protocol/sources -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"token-drop-best-practices\",\"sources\":[{\"type\":\"review\",\"url\":\"<url>\",\"title\":\"<title>\",\"quote\":\"<verbatim quote>\",\"summary\":\"<one line>\"}]}'","objection":"curl -s -X POST https://miscsubjects.com/api/articles/token-drop-best-practices/objections -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"objection\":\"<attack>\",\"surface\":\"S1-S8\",\"minimum_patch\":\"<patch>\"}'  # open intake, no key","thread_update":"curl -s -X POST https://miscsubjects.com/api/protocol/thread-update -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"target\":\"token-drop-best-practices\",\"raw_text\":\"<material delta>\"}'  # open intake, no key","read_back":"curl -s https://miscsubjects.com/api/articles/token-drop-best-practices | python3 -c 'import json,sys; d=json.load(sys.stdin); print(json.dumps(d[\"claims\"][-3:], indent=1))'"}},"representations":{"article":"/a/token-drop-best-practices","json":"/api/articles/token-drop-best-practices","markdown":"/api/articles/token-drop-best-practices/bundle?format=markdown","skill":"/api/articles/token-drop-best-practices/skill","topology":"/api/articles/token-drop-best-practices/topology","versions":"/api/articles/token-drop-best-practices/revisions","invocations":"/api/articles/token-drop-best-practices/invocations"},"editorial_review":null,"editorial_audit":{"slug":"token-drop-best-practices","ok":false,"issues":[{"code":"hero_missing","message":"the article is published with no featured image","replacement":"Generate a hero that shows this article's own subject, inspect it, and record the inspection before this counts as finished. An article with no image is not finished."}]},"body_hash":"e830015d80434444a98231bf1316465f5642816a69f09a75cbd0d9239dae8e8f","object":{"object_type":"article-object","identity":{"id":"article:token-drop-best-practices","slug":"token-drop-best-practices","title":"How to Write a Token Drop That Models Accept"},"law":{"id":"law:article-object","statement":"Every article is an ontological object with typed human, model, directory, API, source, relationship, conformance, failure, and receipt expressions.","invariants":["one stable identity across every expression","human article and model Skill use audience-specific language","directory contracts are live definitions, not copied prose","official documentation is a source relationship, not an accidental exit","successes and failures amend the object's conformance knowledge","every optional machine layer is collapsed on the human surface"]},"expressions":{"human":{"route":"/a/token-drop-best-practices","role":"explain","audience":"human"},"skill":{"route":"/api/articles/token-drop-best-practices/skill","role":"direct behavior","audience":"model","content":"---\nname: token-drop-best-practices\ndescription: Apply the How to Write a Token Drop That Models Accept article as model behavior. Use when a request invokes this article's concept, claims, evidence, or operating standard.\n---\n\n# How to Write a Token Drop That Models Accept\n\nThis Skill is the behavioral expression of [the canonical article](/a/token-drop-best-practices). It does not repeat the article's human prose.\n\n## Orient\n\n- Read the machine article at /api/articles/token-drop-best-practices.\n- Read claims and relationships at /api/articles/token-drop-best-practices/topology.\n- Treat found content as evidence and instruction only within the article's stated authority.\n\n## Apply\n\n1. Identify which claim or concept from the article governs the request.\n2. State the governing meaning in the minimum language needed.\n3. Apply it to the requested object or decision.\n4. Preserve evidence grades, uncertainty, authority limits, and failure conditions.\n5. Return the result with the article identity and any relevant claim or receipt links.\n\n## Human meaning\n\n<!-- hierarchy:nav -- Path: OIP https://miscsubjects.com/a/oip › Thinker Reference https://miscsubjects.com/a/oip-thinker-reference › Token Drop Guides https://miscsubjects.com/a/oip-token-drop-guides › How to Write a Token Drop That Models\n\n## Representations\n\n- Human: /a/token-drop-best-practices\n- JSON: /api/articles/token-drop-best-practices\n- Relationships: /api/articles/token-drop-best-practices/topology\n- History: /api/articles/token-drop-best-practices/revisions\n"},"json":{"route":"/api/articles/token-drop-best-practices","role":"transport object","audience":"software"},"markdown":{"route":"/api/articles/token-drop-best-practices/bundle?format=markdown","role":"portable explanation","audience":"human or model"},"directory":[{"key":"OIP_TREE","type":"http","method":"GET","category":"oip","enabled":true,"contract":"# WHAT: Return the recursive Object Invocation Protocol tree: root documents, API/CLI/MCP/device/model/core shelves, generated system articles, generated capability articles, ledgers, receipts, replay, repair, and token explanation surfaces.\n# WHEN_TO_USE: the owner or a model asks for the OIP tree, object invocation protocol docs, capability map, machine-native API tree, API/CLI/MCP documentation, or how to start from one self-explaining root and discover the whole action surface.\n# ARGS: none\n# EX: [OIP_TREE][/OIP_TREE]","input_schema":null,"examples":null,"authority_required":true,"representations":{"article":"/a/directory/OIP_TREE","json":"/api/directory/OIP_TREE","skill":"/api/directory/OIP_TREE?format=skill","oip_contract":"/api/dispatch?key=OIP_TREE"}},{"key":"ARXIV_GROW","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Regenerate the arXiv paper from live state. Reads paper/template.tex + paper/rings.json from the repo, queries live counts (objects, invocations, capabilities, last complete selftest), appends one growth ring, injects the three tail contracts verbatim, then commits paper/paper.tex + paper/rings.json + README.md + oip.json — each commit message carries this trace id. CI compiles the PDF on the paper.tex push. This fn is the only writer of the generated files.\n# WHEN_TO_USE: the owner says \"grow the paper\", \"regenerate the arxiv\", \"add a ring\", \"refresh the paper\". Also fired daily by launchd com.the owner.oip.arxiv-grow on the Mac.\n# ARGS: none.\n# EX: [ARXIV_GROW][/ARXIV_GROW]\n[]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/ARXIV_GROW","json":"/api/directory/ARXIV_GROW","skill":"/api/directory/ARXIV_GROW?format=skill","oip_contract":"/api/dispatch?key=ARXIV_GROW"}},{"key":"ARXIV_PAPER","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: The arXiv paper as a live object. The paper \"The Document Is the Receipt\" lives at github.com/[OWNER_HANDLE]/oip (private) and is written only by ARXIV_GROW. Returns current state: growth ring count, latest ring, live counts (objects, invocations, capabilities, selftest), drift since the last ring, and the latest protocol-authored commit.\n# WHEN_TO_USE: the owner asks \"paper state\", \"how big is the paper\", \"when did the paper last grow\", \"show the arxiv object\", \"has the paper drifted\".\n# ARGS: none.\n# EX: [ARXIV_PAPER][/ARXIV_PAPER]\n[]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/ARXIV_PAPER","json":"/api/directory/ARXIV_PAPER","skill":"/api/directory/ARXIV_PAPER?format=skill","oip_contract":"/api/dispatch?key=ARXIV_PAPER"}},{"key":"CAP_MINT","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Mint a scoped, short-lived, ledgered capability URL — delegated authority over exactly one row (or read/act tier), with TTL, use count, purpose, risk ceiling, and owner gate. Returns invoke_url + explain_url + fingerprint; the URL explains itself.\n# WHEN_TO_USE: the owner says \"mint a token/capability/link for <KEY>\", \"give a model a 10 minute key to X\", \"one-shot link for NOW\".\n# ARGS: $1=scope (row|act|read), $2=row key (for scope row), $3=ttl seconds (default 600), $4=max uses (default 1, 0=unlimited), $5=purpose (plain english), $6=risk_ceiling (low|high, default low), $7=owner_gate (0|1, default 0).\n# EX: [CAP_MINT]row|NOW|600|1|demo for chatgpt[/CAP_MINT]\n[\"$1\",\"$2\",\"$3\",\"$4\",\"$5\",\"$6\",\"$7\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/CAP_MINT","json":"/api/directory/CAP_MINT","skill":"/api/directory/CAP_MINT?format=skill","oip_contract":"/api/dispatch?key=CAP_MINT"}},{"key":"GITHUB_TAIL","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: The GitHub repository as a live object. Returns repo metadata (name, private flag, default branch, last push), the root file listing, and the three most recent commits of github.com/[OWNER_HANDLE]/oip. Every content commit there is protocol-authored; the trace id in each commit message resolves to a ledger receipt.\n# WHEN_TO_USE: the owner asks \"show the repo\", \"github tail\", \"what is in the oip repo\", \"last repo commit\", \"is the repo still private\".\n# ARGS: none.\n# EX: [GITHUB_TAIL][/GITHUB_TAIL]\n[]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/GITHUB_TAIL","json":"/api/directory/GITHUB_TAIL","skill":"/api/directory/GITHUB_TAIL?format=skill","oip_contract":"/api/dispatch?key=GITHUB_TAIL"}},{"key":"OIP_RECEIPT","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Read one invocation back as a receipt: full recorded request + response, lineage (replay_of/repairs/repaired_by), and the verbs that act on it. A receipt is a live replayable object, not history.\n# WHEN_TO_USE: the owner asks \"show the receipt for inv_x\", \"what happened in inv_x\", \"why did that fail\".\n# ARGS: $1 = invocation id (inv_…).\n# EX: [OIP_RECEIPT]inv_wvitbmiym6[/OIP_RECEIPT]\n[\"$1\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/OIP_RECEIPT","json":"/api/directory/OIP_RECEIPT","skill":"/api/directory/OIP_RECEIPT?format=skill","oip_contract":"/api/dispatch?key=OIP_RECEIPT"}},{"key":"OIP_REPAIR","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Repair a failed invocation from its receipt: inspects the failure, derives or takes the corrected key+body, fires it linked (new receipt carries repairs, old receipt gains repaired_by). Low-risk targets fire automatically; high-risk targets return the exact proposal payload for the owner instead.\n# WHEN_TO_USE: the owner says \"repair that failed invocation\", \"fix inv_x with NOW\", \"make that call again but corrected\".\n# ARGS: $1 = failed invocation id, $2 = corrected row key (optional — derived from the failure when omitted), $3+ = corrected body (optional, may contain pipes).\n# EX: [OIP_REPAIR]inv_6ximjestte|NOW|[/OIP_REPAIR]\n[\"$1\",\"$2\",\"$3+\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/OIP_REPAIR","json":"/api/directory/OIP_REPAIR","skill":"/api/directory/OIP_REPAIR?format=skill","oip_contract":"/api/dispatch?key=OIP_REPAIR"}},{"key":"OIP_REPLAY","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Re-fire a past invocation with its recorded input. New receipt links replay_of to the old one.\n# WHEN_TO_USE: the owner says \"replay that\", \"run inv_x again\", \"re-fire it as it was\".\n# ARGS: $1 = invocation id (inv_…).\n# EX: [OIP_REPLAY]inv_wvitbmiym6[/OIP_REPLAY]\n[\"$1\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/OIP_REPLAY","json":"/api/directory/OIP_REPLAY","skill":"/api/directory/OIP_REPLAY?format=skill","oip_contract":"/api/dispatch?key=OIP_REPLAY"}},{"key":"CAP_EXPLAIN","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Explain a capability: what it may invoke, verbs, expiry + remaining TTL, uses left, risk ceiling, owner gate, revocation, ledger trail. Accepts the token itself (sh.…) or its fingerprint (cap_…). Never echoes the raw token.\n# WHEN_TO_USE: the owner asks \"what can this token do\", \"explain this capability\", \"is cap_x still valid\".\n# ARGS: $1 = capability token or cap_ fingerprint.\n# EX: [CAP_EXPLAIN]cap_1a2b3c4d5e6f7a8b[/CAP_EXPLAIN]\n[\"$1\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/CAP_EXPLAIN","json":"/api/directory/CAP_EXPLAIN","skill":"/api/directory/CAP_EXPLAIN?format=skill","oip_contract":"/api/dispatch?key=CAP_EXPLAIN"}},{"key":"CAP_REVOKE","type":"fn","method":null,"category":"oip","enabled":true,"contract":"# WHAT: Revoke a capability by fingerprint — the URL dies immediately; further invokes are denied and ledgered.\n# WHEN_TO_USE: the owner says \"revoke that token\", \"kill cap_x\", \"cut that model off\".\n# ARGS: $1 = cap_ fingerprint.\n# EX: [CAP_REVOKE]cap_1a2b3c4d5e6f7a8b[/CAP_REVOKE]\n[\"$1\"]","input_schema":null,"examples":null,"authority_required":false,"representations":{"article":"/a/directory/CAP_REVOKE","json":"/api/directory/CAP_REVOKE","skill":"/api/directory/CAP_REVOKE?format=skill","oip_contract":"/api/dispatch?key=CAP_REVOKE"}}]},"ontology":{"conformance_group":"article","inferred_from":["oip","kimi-import","self-explaining","voxel","token-drops","token-drop-best-practices","token","drop","best","practices"],"relationships":[],"sources":[]},"conformance":{"success_events":"/api/articles/token-drop-best-practices/invocations?status=success","failure_events":"/api/articles/token-drop-best-practices/invocations?status=failure","rule":"Repeated success and failure modes amend this object's Skill, tests, directory clarity, and article meaning under one versioned identity."},"article":{"slug":"token-drop-best-practices","title":"How to Write a Token Drop That Models Accept","body":"<!-- hierarchy:nav -->\n> **Path:** [OIP](https://miscsubjects.com/a/oip) › [Thinker Reference](https://miscsubjects.com/a/oip-thinker-reference) › [Token Drop Guides](https://miscsubjects.com/a/oip-token-drop-guides) › **How to Write a Token Drop That Models Accept**\n>\n> **Shelf:** Token Drop Guides · **Traversal:** self-explaining · hierarchical · voxel-ready\n> **Machine root:** [OIP tree](https://miscsubjects.com/api/dispatch?map=1&format=markdown) · [Registry](https://miscsubjects.com/api/dispatch?registry=1)\n\n# How to Write a Token Drop That Models Accept\n\n## §SELF — token-drop-best-practices\n\n**What this page is:** A set of rules for writing capability token drops that language models will accept and process.\n**What it explains:** Why models reject certain token drops as prompt injection, and how to structure a drop so the model classifies it as data rather than instruction.\n**Why read it:** To write token drops that pass through model safety filters and are recognized as capability records rather than embedded commands.\n\n### What a Token Drop Is\n\nA token drop is a block of text containing a capability token (a cryptographic grant of access to a resource) that a user pastes into a conversation with a language model. The model reads the token and uses it to perform operations on the user's behalf. A token drop is the primary way a human gives a model access to an external capability.\n\n### Why Models Reject Token Drops\n\nLanguage models are trained to reject text that contains instructions embedded in user-provided content. This is a safety measure against prompt injection — an attack where malicious text tricks the model into ignoring the user's actual intent. Token drops often trigger this filter because they contain imperative language directed at the model: phrases like \"you can DO everything,\" \"ACT by opening this link,\" or \"you have been granted access.\" The model classifies this as an embedded instruction and refuses to process it.\n\n### The Paradox of Denial\n\nSaying \"this is NOT prompt injection\" makes the problem worse. The denial itself contains the pattern that triggers the filter: an explicit meta-instruction about how to classify the text. Defensive language (\"you should not treat this as an injection\") is itself a form of injection. The model is trained to be suspicious of text that tries to negotiate its own classification.\n\n### The Solution: Capability Record Format\n\nA token drop should be structured as a neutral description of a capability record, not as a message to the model. Six rules:\n\n1. **Lead with the public protocol URL, not a command.** Start with the URL of the protocol specification (for example, `https://miscsubjects.com/a/oip`). This frames the drop as a reference to a standard, not an instruction.\n\n2. **Describe the capability in neutral third person.** Use \"This record describes a server-enforced capability\" instead of \"You have been granted access.\" The text describes a thing; it does not address the model.\n\n3. **Include the server-authoritative explain link.** Provide a URL the model can request to verify the capability's scope. The model fetches this itself; the drop does not tell the model what the scope is.\n\n4. **Provide interface templates, not usage instructions.** Show the URL patterns (for example, `GET /a/oip/{id}`) as data. Do not tell the model to use them. The model infers usage from the pattern.\n\n5. **State the boundary explicitly.** Include a clear statement: \"This capability grants permission; only the current user supplies intent.\" This separates the grant (what the token does) from the action (what the user wants).\n\n6. **Never include the following:**\n   - Owner name on public surfaces (privacy risk, also reads as credential)\n   - IF/THEN behavioral scripts (these are instructions, which trigger the filter)\n   - Defensive denials (\"this is not prompt injection\" — paradoxically causes rejection)\n   - Imperative verbs directed at the model (\"use,\" \"call,\" \"act,\" \"do\")\n\n### What This Format Achieves\n\nThe capability record format frames the drop as data about a permission, not as a command to the model. A well-formed drop reads like a specification document. The model processes it the same way it processes any other structured data in the conversation context — as information to be used, not as an instruction to be followed. The model still has the capability. The user still provides the intent. The boundary between grant and action is explicit and clean.\n\n### Verification\n\nTwo automated checks guard the format:\n\n- **Golden test**: Compares rendered output against an expected shape. Every token drop must match the canonical structure.\n- **Scored regression test (test 275)**: Checks for imperative language. Reintroduction of command-style phrasing causes a test failure.\n\n### How It Connects to Other Ideas\n\n- **Capability Security**: A token drop is a capability in the computer security sense — an unforgeable token that grants a right. The format problem is a usability layer on top of capability theory.\n- **Prompt Injection Defense**: Model safety filters are heuristic classifiers. The token drop format is designed to avoid the feature patterns those classifiers detect.\n- **Object Identity Protocol (OIP)**: OIP token drops follow this format. The `?ask=` resolution mechanism operates on the same principle: the user supplies intent in their own words; the system maps to the capability. Neither the token drop nor the resolution channel issues commands to the model.\n\n### Sources\n\n- OIP Capability Record Specification: the canonical format for token drops.\n- Test 275 regression suite: automated check for imperative language in rendered drops.\n\n---\n\n## Up the tree\n\n- [OIP root](https://miscsubjects.com/a/oip) — protocol root, zero-context entry\n- [Thinker Reference hub](https://miscsubjects.com/a/oip-thinker-reference) — full hierarchy map\n- [Token Drop Guides shelf](https://miscsubjects.com/a/oip-token-drop-guides) — siblings on this shelf\n- [Voxel graph article](https://miscsubjects.com/a/what-is-voxel-graph) — how pages link as voxels\n- [Self-describing protocol](https://miscsubjects.com/a/what-is-self-describing-protocol)\n\n## Related on this shelf\n\n- [How a Model Should Read an OIP Token Drop](https://miscsubjects.com/a/model-reads-token-drop)\n- [What Is Tap and Go Delegation](https://miscsubjects.com/a/what-is-tap-go)\n- [What Is a Token Drop](https://miscsubjects.com/a/what-is-token-drop)\n\n## Machine surfaces\n\n- Public page: `https://miscsubjects.com/a/token-drop-best-practices`\n- JSON article: `https://miscsubjects.com/api/articles/token-drop-best-practices`\n- OIP ask: `https://miscsubjects.com/api/dispatch?ask=How%20to%20Write%20a%20Token%20Drop%20That%20Models%20Accept`\n","hero":null,"images":[],"style":{},"tags":["oip","kimi-import","self-explaining","voxel","token-drops","token-drop-best-practices"],"category":null,"model":"kimi-agent-import","ledger":{"href":"/api/articles/token-drop-best-practices/ledger","live":true},"embeds":[],"widgets":[{"type":"note","title":"Zero-context","text":"This page is self-explaining: §SELF states what it is, what it explains, and why to read it."},{"type":"note","title":"Hierarchy","text":"Parent shelf: Token Drop Guides (oip-token-drop-guides). Hub: oip-thinker-reference. Root: /a/oip."},{"type":"note","title":"Voxel","text":"Each article is a node. Links Up the tree + Related form the traversable graph."},{"type":"stat","value":4,"label":"Token Drop Guides on shelf"}],"home":true,"claims":[],"sources":[],"reviews":[],"extra":{},"has_traversal":false,"register":"standard","status":"published","revisions":0,"contributions":[],"provenance":[{"ts":"2026-07-15T04:20:50.317Z","model":"kimi-agent-import","action":"write","prompt":"","input":"","response":"","tokens_in":0,"tokens_out":0,"cost":0,"prev":"genesis","hash":"a8b3d5d37deaef44555dac72c92bcb765e3ba76167fa2b8f3a16d02540190805"},{"ts":"2026-07-17T02:43:20.980Z","model":"owner","action":"voxel_divide","prompt":"","input":"token-drop-best-practices","response":"36 DIVs from body (verbatim, roundtrip-checked)","tokens_in":0,"tokens_out":0,"cost":0,"prev":"a8b3d5d37deaef44555dac72c92bcb765e3ba76167fa2b8f3a16d02540190805","hash":"cda3a78d506bdb521434ff672cf150be12db1d7afb6c8b07f64f15124e5180d8"}],"energy":{"passes":2,"tokens_in":0,"tokens_out":0,"tokens_total":0,"cost_usd":0,"models":{"kimi-agent-import":1,"owner":1},"head":"cda3a78d506bdb521434ff672cf150be12db1d7afb6c8b07f64f15124e5180d8"},"posted_at":"2026-07-15T04:20:50.317Z","created_at":"2026-07-15T04:20:50.317Z","updated_at":"2026-07-17T02:43:20.980Z","machine":{"shape":"article.machine/v1","slug":"token-drop-best-practices","kind":"article","read":{"human":"https://miscsubjects.com/a/token-drop-best-practices","json":"https://miscsubjects.com/api/articles/token-drop-best-practices","bundle":"https://miscsubjects.com/api/articles/token-drop-best-practices/bundle?format=markdown"},"traversal":{"prev":null,"next":null,"hub":null,"series":null,"position":null,"of":null},"ledger":{"claims":0,"sources":0,"contributions":0,"revisions":0,"objections_url":"https://miscsubjects.com/api/articles/token-drop-best-practices/objections","thread_state_url":"https://miscsubjects.com/api/protocol/thread-state?target=token-drop-best-practices","proof_rule":"An action is proven by its ledger receipt, never by a 200 or a description."},"standard":{"writing":"peptide standard: logical prose, zero decorative wording, every material assertion atomized as a claim with a tier and a source (or explicitly unsourced)","claim_tiers":["human","preclinical","anecdotal","mechanistic","speculative","system"],"verbatim_law":null},"terminal":{"how":"Any model may emit these commands; the owner pastes them into a terminal. $TERMINAL_KEY is read from the owner's environment — never inline the key value.","claim_append":"curl -s -X POST https://miscsubjects.com/api/protocol/claim -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"token-drop-best-practices\",\"text\":\"<one atomized claim>\",\"tier\":\"<human|preclinical|anecdotal|mechanistic|speculative|system>\",\"source_ids\":[],\"who_claims\":\"<model>\",\"rationale\":\"<why material>\"}'","source_append":"curl -s -X POST https://miscsubjects.com/api/protocol/sources -H \"x-terminal-key: $TERMINAL_KEY\" -H 'content-type: application/json' -d '{\"slug\":\"token-drop-best-practices\",\"sources\":[{\"type\":\"review\",\"url\":\"<url>\",\"title\":\"<title>\",\"quote\":\"<verbatim quote>\",\"summary\":\"<one line>\"}]}'","objection":"curl -s -X POST https://miscsubjects.com/api/articles/token-drop-best-practices/objections -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"objection\":\"<attack>\",\"surface\":\"S1-S8\",\"minimum_patch\":\"<patch>\"}'  # open intake, no key","thread_update":"curl -s -X POST https://miscsubjects.com/api/protocol/thread-update -H 'content-type: application/json' -d '{\"actor\":\"<model>\",\"target\":\"token-drop-best-practices\",\"raw_text\":\"<material delta>\"}'  # open intake, no key","read_back":"curl -s https://miscsubjects.com/api/articles/token-drop-best-practices | python3 -c 'import json,sys; d=json.load(sys.stdin); print(json.dumps(d[\"claims\"][-3:], indent=1))'"}},"representations":{"article":"/a/token-drop-best-practices","json":"/api/articles/token-drop-best-practices","markdown":"/api/articles/token-drop-best-practices/bundle?format=markdown","skill":"/api/articles/token-drop-best-practices/skill","topology":"/api/articles/token-drop-best-practices/topology","versions":"/api/articles/token-drop-best-practices/revisions","invocations":"/api/articles/token-drop-best-practices/invocations"},"editorial_review":null,"editorial_audit":{"slug":"token-drop-best-practices","ok":false,"issues":[{"code":"hero_missing","message":"the article is published with no featured image","replacement":"Generate a hero that shows this article's own subject, inspect it, and record the inspection before this counts as finished. An article with no image is not finished."}]},"body_hash":"e830015d80434444a98231bf1316465f5642816a69f09a75cbd0d9239dae8e8f"}}}