fix(ipfs2filecoin): design review follow-ups for the landing page - #350
Merged
SgtPooki merged 13 commits intoAug 14, 2026
Merged
Conversation
Faq hardcoded divide-white/20, and Accordion hardcoded text-zinc-200 on the chevron and text-zinc-300 on the body. All three assume a dark background, which held while the homepage was the only consumer since it mounts the FAQ on a dark section. On a light or gray section the dividers and chevrons are effectively invisible. They now use --color-border-base, --color-text-base and --color-paragraph-text, which flip off the .light-section / .dark-section class that Section already sets, so no isDark branching is needed. Faq keeps useBackground purely for prose-invert. Moving the colour onto the item rather than the parent also fixes the first divider. divide-* targets every child after the first, so item one's border was falling back to currentColor and rendering at full brightness on dark. The homepage FAQ shifts slightly as a result: chevron zinc-200 to zinc-50, body zinc-300 to zinc-400, both now matching the tokens used everywhere else. That page is worth a look alongside this one. Review: @filipagr Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Follow-ups from a walkthrough of the page with Filipa Ribeiro. Alignment and spacing - "Estimate your deposit" and the hero sat 60px inset from their siblings, because PageSection already wraps children in a Container and page.tsx nested a second one, doubling the horizontal padding. Removed both, plus the now-unused import. - The Unit and Funded-for selects had the native arrow jammed against the border with a cavern between it and the value. A local SelectField goes appearance-none with a Phosphor caret inset to match the field's text, and the grid drops from three equal columns to 2fr/1fr/2fr so Unit is no longer a full-width field holding three characters. - "What this does not do" is constrained to max-w-5xl; lines ran past 15 words at desktop. Type - Both tables move off text-sm to text-base, with service names at text-lg. - Monospace figures dropped for the body font with tabular-nums. The reported problem was numbers appearing larger than the names beside them, which was a typeface illusion rather than a size difference, so matching the typeface removes it at the root while tabular-nums keeps the columns aligned. Flagging this one as the most debatable call in the batch. - The highlighted Filecoin row carries weight instead of a marginally darker grey, which read as a mistake rather than emphasis. Visual interest - How it works renders 2x2 rather than 3-up with an orphaned fourth, and step numbers sit above each title in the accent colour. This needs a new StepList component because the shared Card types title as a plain string, so the number could only ever be inlined as "01. Hand over the list". - Reason cards take icons, following how agents/ and warm-storage-service/ already pass Phosphor icons to Card. - The agent column takes two thirds and "Talk to us" becomes a one third call-out, giving the path with more to say more room. The external-link arrow beside Filebase and Pinata is tightened with a local override. It is sized for button text inside ExternalTextLink in ui-filecoin, so the real fix belongs upstream; this is a commented stopgap. Still open from the review and deliberately not attempted here: table whitespace now that the type sizes are settled, the hero background treatment, and the optional illustration beside the agent heading. Review: @filipagr Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
parseCidList already validated, deduped and stripped gateway prefixes, but the verdict box reported only a count. The successful path gave you nothing to act on, so the top of the page had no purpose: you still assembled cids.txt by hand. The prompt was in fact offered only on the over-cap branch, so the one outcome that succeeded was the one outcome with no output. A checked list now produces the agent prompt with your own CIDs inlined, via buildAgentPrompt, plus a cids.txt download of the cleaned and deduped list. AGENT_PROMPT stays as the fallback for the empty and over-cap cases, where pointing at a file is the right shape rather than inlining hundreds of lines. The instruction to "estimate what they cost to store below" is replaced. A check reads the CIDs and not the bytes behind them, so it cannot price anything on its own, and implying otherwise was the most confusing part of the flow. Resolving a CID to its actual size needs backend capability and is a separate conversation. Hero polish in the same component: "one per line" sits beside the label in the accent colour rather than surfacing only after a failed check, the button is centred, the empty textarea is shorter, and the free-check facts read as a scannable list instead of fine print. Copy prompt demotes from a full-width button to an icon in the code block's corner. Verified against deliberately messy input: a bare CID, a gateway URL, a duplicate and a junk line resolve to three unique CIDs, with the URL prefix stripped and both the skipped line and the duplicate reported. Review: @filipagr Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
CLAUDE.md is per-machine guidance for Claude Code, and todo.md is the working list of design review actions for /ipfs2filecoin. Neither belongs in the repo. Note that this keeps todo.md local, so it does not travel to reviewers via git. Review: @filipagr Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
@gmoranxyz is attempting to deploy a commit to the FilOz Team on Vercel. A member of the Team first needs to authorize it. |
Contributor
Author
|
@filipagr take a look and let me know what you think as a first pass! |
Fold the label, textarea and action into a single bordered panel with a live "X of Y lines are CIDs" readout and a Cmd/Ctrl+Enter shortcut. Move the verdict into its own CidListVerdict component with an icon per outcome, and extract pluralize so the readout and verdict agree. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… responsive Give each pricing row a detail sub-line so the provider is what you compare on and the plan tier stays secondary, and key rows by provider plus tier now that two Pinata tiers share a name. On mobile the three columns do not fit, so each provider becomes a stacked card; the full table returns at md. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… hero and type Move the estimator into a companion card beside the comparison table (the table is the market rate, the card prices it for your data), with the two inputs side by side and a compact stacked result that fits the narrow column. Add a faint brand glow along the bottom of the hero, step the reason-card and agent-warning copy down a size, and render How it works four across on desktop. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
fix(ipfs2filecoin): estimator beside the table, responsive pricing, hero polish
…COPY The rate footnote, one reason card, and the SEO description each restated $2.50 as a literal, so they could silently drift from the comparison table and estimator, which already derive from the constant.
fix(ipfs2filecoin): derive the $2.50 rate from USD_PER_TIB_MONTH_PER_COPY
gmoranxyz
marked this pull request as ready for review
August 5, 2026 14:30
Contributor
Author
|
@jennijuju I think this is ready for you to take a look! Please forgive anything we broke with our clumsy vibe coding! |
SgtPooki
approved these changes
Aug 14, 2026
github-project-automation
Bot
moved this from 📌 Triage
to ✔️ Approved by reviewer
in FOC
SgtPooki
merged commit Aug 14, 2026
4b57966
into
FilOzone:feat/ipfs2filecoin-landing-page
5 checks passed
SgtPooki
added a commit
that referenced
this pull request
Sep 1, 2026
* fix(faq): make accordion dividers and chevrons background-aware Faq hardcoded divide-white/20, and Accordion hardcoded text-zinc-200 on the chevron and text-zinc-300 on the body. All three assume a dark background, which held while the homepage was the only consumer since it mounts the FAQ on a dark section. On a light or gray section the dividers and chevrons are effectively invisible. They now use --color-border-base, --color-text-base and --color-paragraph-text, which flip off the .light-section / .dark-section class that Section already sets, so no isDark branching is needed. Faq keeps useBackground purely for prose-invert. Moving the colour onto the item rather than the parent also fixes the first divider. divide-* targets every child after the first, so item one's border was falling back to currentColor and rendering at full brightness on dark. The homepage FAQ shifts slightly as a result: chevron zinc-200 to zinc-50, body zinc-300 to zinc-400, both now matching the tokens used everywhere else. That page is worth a look alongside this one. Review: @filipagr * fix(ipfs2filecoin): correct layout and type hierarchy from design review Follow-ups from a walkthrough of the page with Filipa Ribeiro. Alignment and spacing - "Estimate your deposit" and the hero sat 60px inset from their siblings, because PageSection already wraps children in a Container and page.tsx nested a second one, doubling the horizontal padding. Removed both, plus the now-unused import. - The Unit and Funded-for selects had the native arrow jammed against the border with a cavern between it and the value. A local SelectField goes appearance-none with a Phosphor caret inset to match the field's text, and the grid drops from three equal columns to 2fr/1fr/2fr so Unit is no longer a full-width field holding three characters. - "What this does not do" is constrained to max-w-5xl; lines ran past 15 words at desktop. Type - Both tables move off text-sm to text-base, with service names at text-lg. - Monospace figures dropped for the body font with tabular-nums. The reported problem was numbers appearing larger than the names beside them, which was a typeface illusion rather than a size difference, so matching the typeface removes it at the root while tabular-nums keeps the columns aligned. Flagging this one as the most debatable call in the batch. - The highlighted Filecoin row carries weight instead of a marginally darker grey, which read as a mistake rather than emphasis. Visual interest - How it works renders 2x2 rather than 3-up with an orphaned fourth, and step numbers sit above each title in the accent colour. This needs a new StepList component because the shared Card types title as a plain string, so the number could only ever be inlined as "01. Hand over the list". - Reason cards take icons, following how agents/ and warm-storage-service/ already pass Phosphor icons to Card. - The agent column takes two thirds and "Talk to us" becomes a one third call-out, giving the path with more to say more room. The external-link arrow beside Filebase and Pinata is tightened with a local override. It is sized for button text inside ExternalTextLink in ui-filecoin, so the real fix belongs upstream; this is a commented stopgap. Still open from the review and deliberately not attempted here: table whitespace now that the type sizes are settled, the hero background treatment, and the optional illustration beside the agent heading. Review: @filipagr * feat(ipfs2filecoin): turn a checked CID list into something actionable parseCidList already validated, deduped and stripped gateway prefixes, but the verdict box reported only a count. The successful path gave you nothing to act on, so the top of the page had no purpose: you still assembled cids.txt by hand. The prompt was in fact offered only on the over-cap branch, so the one outcome that succeeded was the one outcome with no output. A checked list now produces the agent prompt with your own CIDs inlined, via buildAgentPrompt, plus a cids.txt download of the cleaned and deduped list. AGENT_PROMPT stays as the fallback for the empty and over-cap cases, where pointing at a file is the right shape rather than inlining hundreds of lines. The instruction to "estimate what they cost to store below" is replaced. A check reads the CIDs and not the bytes behind them, so it cannot price anything on its own, and implying otherwise was the most confusing part of the flow. Resolving a CID to its actual size needs backend capability and is a separate conversation. Hero polish in the same component: "one per line" sits beside the label in the accent colour rather than surfacing only after a failed check, the button is centred, the empty textarea is shorter, and the free-check facts read as a scannable list instead of fine print. Copy prompt demotes from a full-width button to an icon in the code block's corner. Verified against deliberately messy input: a bare CID, a gateway URL, a duplicate and a junk line resolve to three unique CIDs, with the URL prefix stripped and both the skipped line and the duplicate reported. Review: @filipagr * chore: ignore local agent and working files CLAUDE.md is per-machine guidance for Claude Code, and todo.md is the working list of design review actions for /ipfs2filecoin. Neither belongs in the repo. Note that this keeps todo.md local, so it does not travel to reviewers via git. Review: @filipagr * feat(ipfs2filecoin): rework the CID checker into one input surface Fold the label, textarea and action into a single bordered panel with a live "X of Y lines are CIDs" readout and a Cmd/Ctrl+Enter shortcut. Move the verdict into its own CidListVerdict component with an icon per outcome, and extract pluralize so the readout and verdict agree. * fix(ipfs2filecoin): split provider from tier and make the price table responsive Give each pricing row a detail sub-line so the provider is what you compare on and the plan tier stays secondary, and key rows by provider plus tier now that two Pinata tiers share a name. On mobile the three columns do not fit, so each provider becomes a stacked card; the full table returns at md. * fix(ipfs2filecoin): pair the deposit estimator with the table, refine hero and type Move the estimator into a companion card beside the comparison table (the table is the market rate, the card prices it for your data), with the two inputs side by side and a compact stacked result that fits the narrow column. Add a faint brand glow along the bottom of the hero, step the reason-card and agent-warning copy down a size, and render How it works four across on desktop. * fix(ipfs2filecoin): derive the $2.50 rate from USD_PER_TIB_MONTH_PER_COPY The rate footnote, one reason card, and the SEO description each restated $2.50 as a literal, so they could silently drift from the comparison table and estimator, which already derive from the constant. --------- Co-authored-by: filipagr <filipagoncalvesribeiro@gmail.com> Co-authored-by: Russell Dempsey <1173416+SgtPooki@users.noreply.github.com>
SgtPooki
added a commit
that referenced
this pull request
Sep 1, 2026
* feat(ipfs2filecoin): add IPFS to Filecoin campaign landing page Adds filecoin.cloud/ipfs2filecoin, the front door for the IPFS-to-Filecoin mini-campaign, ported from Jenni's wireframe in FilOzone/FOC-PM onto the site's existing ui-filecoin components. The page is both the campaign's front door and its measuring instrument. It does three things: takes a pasted CID list, produces a deposit estimate from a declared volume, and routes to the agent door or to contact. Interactive parts run entirely client-side, with no wallet and no network calls, so a check costs the visitor nothing and sends us nothing: - CID list checker: validates, dedupes, and steers lists over the 500-item browser cap to the agent prompt. - Deposit estimator: reproduces the PRD's worked example exactly ($128.08 deposit, $5.25 refundable for 1 TiB over two years) from documented constants rather than hardcoded figures. - Agent prompt with copy-to-clipboard, the primary conversion. Plausible events cover the funnel the PRD asks for: CID List Checked, Estimate Viewed, and Agent Prompt Copied, the last being the headline number. Also serves a draft migration brief at /ipfs2filecoin/llms.txt and advertises it via <link rel="alternate" type="text/markdown">, so an agent pointed at the human URL discovers the brief without being told the convention. The brief opens with a stage 0 ingress check, per the 2026-07-28 validation run that found runs die there after the expensive commP pass. The page is unlinked from site navigation, reachable by URL and sitemap only, so campaign attribution stays clean. * feat(ipfs2filecoin): replace the draft brief with an executable runbook A real agent run against the draft brief failed at the first hop: the brief never named the tool, so the agent correctly refused to improvise a pipeline and stopped. It also read the 500 GiB figure as a self-serve cap and routed a 56k-CID job to the contact form, which killed the conversion outright. Both are fixed, and the format question behind them is settled. llms.txt was the wrong format. Per llmstxt.org it is a documentation index -- H1, blockquote, then H2 lists of links -- so a procedure does not belong in it. The procedure now lives at /ipfs2filecoin/migrate.md, named for what it is, and llms.txt is used correctly in two places: a spec-conformant site index at /llms.txt, and a campaign index at /ipfs2filecoin/llms.txt whose first link is the runbook. Discovery never depended on the filename anyway -- the prompt carries the URL, and the page carries rel="alternate". The runbook names ipfs2foc, pins the version its flags were verified against, and carries the real command sequence read out of that repo. Notable content: - One invocation form, stated once. A bare `ipfs2foc` after only an `npx` check was the most likely first failure. - Prerequisites split into "needed now" and "needed before stage 4", so a missing key no longer blocks the free stages. - Stage 0 settles ingress before the commitment pass, and records that the relay only serves the single-asset path. - Stage 2 reports from analyze --json using fields that actually exist, with explicit cost and ETA formulas so nothing is invented. - Stage 4 is executed by the user in their own terminal. The agent is told never to accept a private key -- a key pasted into a chat is a key in a transcript. - Stage 5 spawns the pull source as a background process and extracts the tunnel URL from its log, rather than saying "terminal A". - Duration is framed as expected, not as a limit: a long-running command is working, not hung, and there is no cap on CID count. Page copy follows: SELF_SERVE_VOLUME_CAP_LABEL becomes COORDINATION_VOLUME_LABEL and reads as coordination rather than a ceiling, so an agent reading the page cannot reproduce the original false blocker. Reviewed by Codex and Gemini via the peer-review skill; every finding above came out of that pass. Cursor could not run -- it needs auth. * fix(ipfs2filecoin): correct two runbook claims the CLI does not support Audited every command and flag in the runbook against the USAGE string in ipfs2foc's packages/cli/src/index.ts, which is authoritative. Two were wrong. `ipfs2foc --version` is not a subcommand. It falls through to the default branch, writes "unknown command: --version" to stderr, and sets exit code 1. As the very first step of the runbook that reads like a broken install, which is exactly the first-hop failure this rewrite exists to remove. The install check is now `--help`, with a note saying why. The provider's minimum piece size is advisory, not enforced. `pdp-submit` warns and proceeds by default and only refuses under `--strict-piece-size` (see submit-pdp.ts: "The advertised floor is advisory in practice"). The runbook claimed items below the minimum fail at pull time, which would have pushed agents onto the multi-asset path unnecessarily -- it needs staging disk and rules out the relay. Now stated as the trade it is: the minimum is a packing efficiency question, the pull limit is the hard one. * chore(ipfs2filecoin): re-verify competitor rates Checked all four rows against filebase.com/pricing and pinata.cloud/pricing. Every rate still matches: Filebase Pro storage $0.015/GB and IPFS egress $0.015/GB; Pinata Fiesta $0.035/GB and $0.080/GB; Pinata Picnic $0.070/GB and $0.100/GB. No figures changed. The verification date lives in the source comment rather than on the page: it is a maintenance note for us, it ages badly in public, and the two pricing links already let a reader check the current numbers. * fix(ipfs2filecoin): stop pinning the CLI to an unpublished version The runbook told agents to run `npm install -g ipfs2foc@0.6.0`. npm has only 0.1.0 through 0.4.0, so that resolves to nothing and fails with ETARGET on the first command of the procedure. 0.6.0 is the in-tree version; the repo runs ahead of what has been released. Installing latest is correct and safe here. I unpacked the published 0.4.0 and checked its bundle directly: all fourteen commands the runbook uses are present (analyze, commp, create-data-set, export, gas, import-manifest, pack-cars, pdp-submit, plan, probe, redirect-serve, report, serve, status), as is every flag it passes, including --strict-piece-size, --ingress, --source-relay, --no-auto-pack, --sample and --car-store. 0.4.0 also has no --version subcommand, so the --help install check holds there too. The frontmatter now carries tool_min_version rather than an exact pin, since what an agent needs to know is the floor, not the version someone happened to verify against. * fix(ipfs2filecoin): add the flat data set fee once, not per TiB * feat(ipfs2filecoin): rewrite the runbook for the direct-upload flow * fix(ipfs2filecoin): design review follow-ups for the landing page (#350) * fix(faq): make accordion dividers and chevrons background-aware Faq hardcoded divide-white/20, and Accordion hardcoded text-zinc-200 on the chevron and text-zinc-300 on the body. All three assume a dark background, which held while the homepage was the only consumer since it mounts the FAQ on a dark section. On a light or gray section the dividers and chevrons are effectively invisible. They now use --color-border-base, --color-text-base and --color-paragraph-text, which flip off the .light-section / .dark-section class that Section already sets, so no isDark branching is needed. Faq keeps useBackground purely for prose-invert. Moving the colour onto the item rather than the parent also fixes the first divider. divide-* targets every child after the first, so item one's border was falling back to currentColor and rendering at full brightness on dark. The homepage FAQ shifts slightly as a result: chevron zinc-200 to zinc-50, body zinc-300 to zinc-400, both now matching the tokens used everywhere else. That page is worth a look alongside this one. Review: @filipagr * fix(ipfs2filecoin): correct layout and type hierarchy from design review Follow-ups from a walkthrough of the page with Filipa Ribeiro. Alignment and spacing - "Estimate your deposit" and the hero sat 60px inset from their siblings, because PageSection already wraps children in a Container and page.tsx nested a second one, doubling the horizontal padding. Removed both, plus the now-unused import. - The Unit and Funded-for selects had the native arrow jammed against the border with a cavern between it and the value. A local SelectField goes appearance-none with a Phosphor caret inset to match the field's text, and the grid drops from three equal columns to 2fr/1fr/2fr so Unit is no longer a full-width field holding three characters. - "What this does not do" is constrained to max-w-5xl; lines ran past 15 words at desktop. Type - Both tables move off text-sm to text-base, with service names at text-lg. - Monospace figures dropped for the body font with tabular-nums. The reported problem was numbers appearing larger than the names beside them, which was a typeface illusion rather than a size difference, so matching the typeface removes it at the root while tabular-nums keeps the columns aligned. Flagging this one as the most debatable call in the batch. - The highlighted Filecoin row carries weight instead of a marginally darker grey, which read as a mistake rather than emphasis. Visual interest - How it works renders 2x2 rather than 3-up with an orphaned fourth, and step numbers sit above each title in the accent colour. This needs a new StepList component because the shared Card types title as a plain string, so the number could only ever be inlined as "01. Hand over the list". - Reason cards take icons, following how agents/ and warm-storage-service/ already pass Phosphor icons to Card. - The agent column takes two thirds and "Talk to us" becomes a one third call-out, giving the path with more to say more room. The external-link arrow beside Filebase and Pinata is tightened with a local override. It is sized for button text inside ExternalTextLink in ui-filecoin, so the real fix belongs upstream; this is a commented stopgap. Still open from the review and deliberately not attempted here: table whitespace now that the type sizes are settled, the hero background treatment, and the optional illustration beside the agent heading. Review: @filipagr * feat(ipfs2filecoin): turn a checked CID list into something actionable parseCidList already validated, deduped and stripped gateway prefixes, but the verdict box reported only a count. The successful path gave you nothing to act on, so the top of the page had no purpose: you still assembled cids.txt by hand. The prompt was in fact offered only on the over-cap branch, so the one outcome that succeeded was the one outcome with no output. A checked list now produces the agent prompt with your own CIDs inlined, via buildAgentPrompt, plus a cids.txt download of the cleaned and deduped list. AGENT_PROMPT stays as the fallback for the empty and over-cap cases, where pointing at a file is the right shape rather than inlining hundreds of lines. The instruction to "estimate what they cost to store below" is replaced. A check reads the CIDs and not the bytes behind them, so it cannot price anything on its own, and implying otherwise was the most confusing part of the flow. Resolving a CID to its actual size needs backend capability and is a separate conversation. Hero polish in the same component: "one per line" sits beside the label in the accent colour rather than surfacing only after a failed check, the button is centred, the empty textarea is shorter, and the free-check facts read as a scannable list instead of fine print. Copy prompt demotes from a full-width button to an icon in the code block's corner. Verified against deliberately messy input: a bare CID, a gateway URL, a duplicate and a junk line resolve to three unique CIDs, with the URL prefix stripped and both the skipped line and the duplicate reported. Review: @filipagr * chore: ignore local agent and working files CLAUDE.md is per-machine guidance for Claude Code, and todo.md is the working list of design review actions for /ipfs2filecoin. Neither belongs in the repo. Note that this keeps todo.md local, so it does not travel to reviewers via git. Review: @filipagr * feat(ipfs2filecoin): rework the CID checker into one input surface Fold the label, textarea and action into a single bordered panel with a live "X of Y lines are CIDs" readout and a Cmd/Ctrl+Enter shortcut. Move the verdict into its own CidListVerdict component with an icon per outcome, and extract pluralize so the readout and verdict agree. * fix(ipfs2filecoin): split provider from tier and make the price table responsive Give each pricing row a detail sub-line so the provider is what you compare on and the plan tier stays secondary, and key rows by provider plus tier now that two Pinata tiers share a name. On mobile the three columns do not fit, so each provider becomes a stacked card; the full table returns at md. * fix(ipfs2filecoin): pair the deposit estimator with the table, refine hero and type Move the estimator into a companion card beside the comparison table (the table is the market rate, the card prices it for your data), with the two inputs side by side and a compact stacked result that fits the narrow column. Add a faint brand glow along the bottom of the hero, step the reason-card and agent-warning copy down a size, and render How it works four across on desktop. * fix(ipfs2filecoin): derive the $2.50 rate from USD_PER_TIB_MONTH_PER_COPY The rate footnote, one reason card, and the SEO description each restated $2.50 as a literal, so they could silently drift from the comparison table and estimator, which already derive from the constant. --------- Co-authored-by: filipagr <filipagoncalvesribeiro@gmail.com> Co-authored-by: Russell Dempsey <1173416+SgtPooki@users.noreply.github.com> * fix(ipfs2filecoin): drop llms.txt links to the hidden service-providers page * fix(ipfs2filecoin): accept subdomain-style gateway URLs in the CID checker * fix(ipfs2filecoin): describe resumability as the CLI state database, not a browser tab * fix(ipfs2filecoin): runbook requires Node 24 per ipfs2foc 0.8.1 * feat(ipfs2filecoin): drop the 500-item browser check cap * feat(ipfs2filecoin): remove llms.txt routes, the runbook at migrate.md is the agent surface * feat(ipfs2filecoin): carry the agent prompt in the runbook itself * perf(ipfs2filecoin): single-pass CID list parse * docs(ipfs2filecoin): steer agents off pre-upgrade data sets * docs(ipfs2filecoin): disclose CLI telemetry in runbook * feat(ipfs2filecoin): state the list stays in the browser * fix(ipfs2filecoin): dedupe trust line with the fact chips * fix(ipfs2filecoin): keep the check button on one line * feat(ipfs2filecoin): apply product-sync feedback * feat(ipfs2filecoin): lead with the agent prompt * fix(ipfs2filecoin): close with the promised actions * chore(ipfs2filecoin): point prompt at the preview * fix(ipfs2filecoin): drop the last testnet mention * docs(ipfs2filecoin): disclose upload totals telemetry * Revert "chore(ipfs2filecoin): point prompt at the preview" This reverts commit 65760e4. * fix(ipfs2filecoin): close dogfooding-run findings From James's 2026-08-28 mainnet run of the page and runbook. The hero prompt gave an agent no input, so it stalled on the first turn asking for CIDs. Over the inline cap the checker's prompt fell back to that same bare line while the copy beside it claimed the prompt read from a file; it now names cids.txt. The FAQ said the agent hands back both funding and signing. The runbook and the hero say it signs. Funding and placing the key are the user's; signing is not. Stage 4 needs a key in an environment an agent's shell does not keep between commands, and nothing documented how to get one there. It now reads the key from a file without resolving it, written with a prompt rather than a literal so it stays out of shell history. Rule 3 stated an invariant that pattern would have broken. Four promises of a manifest had no referent: nothing writes one. Replaced with the artifacts the run actually produces. The flat per-data-set fee is 0.024 and rendered as $0.02, understating it by up to 17% below ~100 GiB. Rates now format to three decimals; totals still fix at two, so the deposit table is unchanged. Volumes under a gibibyte could not be entered at all. * docs(ipfs2filecoin): tighten the runbook's key rule State the permitted move first. A rule built only of prohibitions puts the forbidden action in front of the agent and never names the one it should take, and the frontmatter's never-list contradicted the key-file pattern stage 3 now documents. Also drops em-dashes from the added comments and prose. * feat(ipfs2filecoin): link the docs migration guide The reciprocal half of the cross-link agreed in the product sync; filecoin-docs#2479 points back here. The runbook is the agent's copy of the migration, so the FAQ about what an agent does is where someone deciding to run it by hand will be standing. The URL resolves once #2479 merges, as ours does for theirs. * docs(ipfs2filecoin): move the runbook to ipfs2foc 0.9.0 0.9.0 is now npm latest, so the two things the runbook worked around are gone. The install check is `--version`, which reports what it installed rather than only proving the binary resolves. Stage 5 reconciles onchain with `report --json`. It used to send the agent to a browser page it could not open, and `report` was unusable anyway: on the upload path it counted its own committed pieces as untracked and exited non-zero on a successful run, so an agent following this document would report a good migration as failed. The explorer link stays for the human. * fix(ipfs2filecoin): drop the wrong data-set upgrade date The runbook told agents a data set created after 2026-08-24 has the gas optimizations. That date is calibration's original target, which itself slipped to 08-31, and mainnet's PDP and FWSS upgrades are not scheduled until 09-03 (filecoin-services#597). The runbook defaults to mainnet, so it was telling agents to reuse mainnet data sets that do not have the optimizations. Fresh sets are already the default and are correct on every network regardless of upgrade timing, so the guidance now rests on that rather than on a date that differs per network and keeps moving. * fix(ipfs2filecoin): deep-link the docs walkthrough The FAQ offers the migration "written out step by step", which is command-line.md. The guide's landing page it pointed at is conceptual, so the link undersold what it promised by one click. --------- Co-authored-by: G <gmoran651@gmail.com> Co-authored-by: filipagr <filipagoncalvesribeiro@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
📝 Description
Design review follow-ups for
/ipfs2filecoin, from a walkthrough of the page with @filipagr on 31 July, plus two rounds of polish since.This targets
feat/ipfs2filecoin-landing-pagerather thanmainon purpose, so it feeds #346 instead of competing with it. Merging here means #346 carries the polish through tomainas one piece.Filipa's overall read was that the structure, content and calls to action are right — opening with "paste your CIDs" instead of a wall of reading is the correct choice. Almost everything below is layout and type hierarchy, plus one genuine functional gap and one drift risk.
Two things to know before reviewing:
src/app/ipfs2filecoin/page.tsx. The base branch has gained four commits this head doesn't have (bc84838runbook rewrite for the direct-upload flow,abbe4caflat data set fee charged once,b67a9fbunpin the CLI,55dcabdre-verify competitor rates), and they touch the same file as the layout rework.CostEstimator.tsx,constants/migration.tsanddata/pricing-comparison.tsall auto-merge; onlypage.tsxneeds hand-resolving.FilOzone:main, and the base is behindmain, so the diff against the base also shows.github/CODEOWNERS(chore: set CODEOWNERS #349), the@filoz/synapse-sdk1.1.1 bump (chore: update @filoz/synapse-sdk to v1.1.1 #348), and the temporary hiding of the Service Providers page with its redirect innext.config.ts(Temporarily hide the Service Providers page #347). Those aren't part of this work and will disappear once the base catches up withmain. Reviewable scope issrc/app/ipfs2filecoin/**plussrc/components/{Accordion,Faq}.tsxand.gitignore.🛠️ Key Changes
Three rounds, each commit scoped to one concern and independently revertable.
Round 1 — design review follow-ups (31 July)
fix(faq)— accordion dividers and chevrons were invisible.Faqhardcodeddivide-white/20andAccordionhardcodedtext-zinc-200/text-zinc-300. All three assume a dark background, which held while the homepage was the only consumer; this page mounts the FAQ on a gray section, where they disappear. Now uses the--color-border-base/--color-text-base/--color-paragraph-texttokens, which already flip off the.light-section/.dark-sectionclassSectionsets — so noisDarkbranching was needed. Moving the colour onto the item rather than the parent also fixes the first divider:divide-*targets every child after the first, so item one's border was falling back tocurrentColorand rendering at full brightness on dark.fix(ipfs2filecoin)— layout and type hierarchy.PageSectionalready wraps children in aContainerandpage.tsxnested a second one, doubling the horizontal padding.SelectFieldgoesappearance-nonewith a Phosphor caret inset to match the field's text.text-smtotext-base, service names totext-lg.tabular-nums.StepList— the sharedCardtypestitleas a plain string, so the number could otherwise only be inlined as "01. Hand over the list".agents/andwarm-storage-service/already pass icons toCard.max-w-5xl— lines ran past 15 words at desktop.feat(ipfs2filecoin)— a checked CID list now produces something.parseCidListalready validated, deduped and stripped gateway prefixes, but the verdict reported only a count, so the top of the page had no purpose — you still builtcids.txtby hand. The prompt was in fact only offered on the over-cap branch, so the one outcome that succeeded was the one with no output. A checked list now yields the agent prompt with your own CIDs inlined viabuildAgentPrompt, plus acids.txtdownload of the cleaned list.AGENT_PROMPTstays as the fallback for the empty and over-cap cases, where pointing at a file is the right shape rather than inlining hundreds of lines. The instruction to "estimate what they cost to store below" is gone: a check reads the CIDs, not the bytes behind them, so it cannot price anything on its own.chore— ignoresCLAUDE.mdand a localtodo.md.Round 2 — @filipagr's polish (#1, 3 August)
CID checker becomes one input surface. Label, textarea and action fold into a single bordered panel with a live "X of Y lines are CIDs" readout and a Cmd/Ctrl+Enter shortcut. The verdict moves into its own
CidListVerdictcomponent with an icon per outcome, andpluralizeis extracted so the readout and the verdict agree.Pricing table compares providers, not plan names. Each row gains a detail sub-line so the provider is what you compare on and the plan tier stays secondary, and rows are keyed by provider plus tier now that two Pinata tiers share a name. Below
mdthe three columns don't fit, so each provider becomes a stacked card; the full table returns atmd.Estimator pairs with the table. It moves into a companion card beside the comparison table — the table is the market rate, the card prices it for your data — at
lg:grid-cols-3with the table on two columns and the estimator on one, its two inputs side by side and a compact stacked result that fits the narrow column. Also: a faint brand glow along the bottom of the hero (--color-brand-500mixed to 12%, so it reads as light on the surface rather than a coloured panel), reason-card and agent-warning copy stepped down a size, and "How it works" four across atlg(2×2 fromsm).Round 3 — kill the hardcoded rate (#2, 5 August)
$2.50was restated as a literal in the rate footnote, one reason card, and the SEO description, so all three could silently drift from the comparison table and estimator, which already derive from the constant. All now go throughformatUsd(USD_PER_TIB_MONTH_PER_COPY).📌 To-Do Before Merging
page.tsxagainst the base's four newer commits. Worth doing as a merge of the base into this branch rather than a rebase, since the history is shared with @filipagr.tabular-numskeeps the columns aligned. Still the most debatable call here, and easy to revert.Accordion/Faqare shared, so the homepage FAQ changes too: chevronzinc-200→zinc-50, bodyzinc-300→zinc-400. Both now match the site's own tokens, but worth eyeballing since it's outside this page's scope.page.tsx. It's sized for button text insideExternalTextLinkinui-filecoin, so the real fix belongs upstream — worth a follow-up issue.Closed since the original review: the hero background treatment, now handled by the brand glow rather than a produced asset; and table whitespace, which the responsive stacked cards and settled type sizes addressed.
Still not attempted, both needing a call rather than code:
🧪 How to Test
nvm use && npm install && npm run dev, then open/ipfs2filecoin.https://ipfs.io/ipfs/…gateway URL, a duplicate, and a junk line. Watch the "X of Y lines are CIDs" readout as you type, then submit with Cmd/Ctrl+Enter rather than the button.lgthe comparison table should occupy two thirds with the "Estimate your deposit" card beside it on one third, tops aligned. Narrow the window: the card drops below the table, and belowmdthe table itself becomes one stacked card per provider.lg, 2×2 fromsm, step numbers above each title in the accent colour.Download cids.txtlink.npm run build(TypeScript, twoslash'd Synapse snippet, all 25 routes) passes at2a8113a, the PR Bump the dependencies group with 11 updates #1 merge. The head has since taken Build Main Components #2 and amainmerge and hasn't been rebuilt locally — the conflict resolution will need a fresh build anyway.📸 Screenshots
Verified in-browser at 1284px. Not attached here — CLI-created PR. Happy to add before/afters for the FAQ, the steps grid, the table/estimator pairing and the CID-check flow if useful.
🔖 Resources
🤖 Generated with Claude Code