B is a substantial, coherent CLI restructuring (forum list/show/post subcommands, mandatory --delegate on posts, updated docs/DSL, RPC next-step strings, and integration tests) that changes real user-facing surface and behavior consistently across code, docs, and tests. A is a smaller but correct fix (min-max gradient instead of ordinal position) with good targeted tests, but it's a narrower, purely cosmetic improvement compared to B's broader architectural and UX rework.
constitution · epochs · watch · epoch 3
comparison
c_66eb04076a98 (tommy-mor) vs c_978e283f2229 (tommy-mor)
download prompt · raw event · cmp_0fa44502dfd95b
council reasoning
B restructures the CLI’s core write path (ingest → `forum post <TAG>` with required delegate), splits forum into explicit list/show/post subcommands, and updates docs, RPC next-move strings, and integration tests—lasting interface design. A only changes rank-row coloring from ordinal position to min–max vote-mass within a group plus unit tests; valuable polish, but much narrower scope.
Side A changes the ranking visualization to derive row colors from each group's actual score range instead of list position, adding a dedicated `score_gradient_t` function, updating rendering to compute per-group min/max scores, and including focused tests for normalization, tied scores, and stability. Side B is a broad CLI and documentation reshaping that mainly renames and reorganizes commands (`ingest` to `forum post`, `forum` to `forum list/show`) and updates help text and tests, providing usability improvements but comparatively less enduring functional value.
sides
A — c_66eb04076a98 (tommy-mor)
message
[0366806e] Color rank rows by vote mass within each group, not list position. Min–max normalization keeps similar scores visually close while still using the full gradient as groups grow and absolute mass dilutes. Co-authored-by: Cursor <cursoragent@cursor.com>
diff preview
diff --git a/server/src/html/mod.rs b/server/src/html/mod.rs
index a58cbbee3490a08a625cb06df06848c59a615d65..4eff2e19ed4d303ff8e80c1eabd8a15b4990e643 100644
--- a/server/src/html/mod.rs
+++ b/server/src/html/mod.rs
@@ -264,12 +264,19 @@ pub fn scope_theme_style(parent: &ItemId) -> String {
)
}
-fn rank_row_style(parent: &ItemId, ordinal: usize, total: usize) -> String {
- let t = if total <= 1 {
- 0.0
- } else {
- ordinal as f64 / (total - 1) as f64
- };
+/// Map vote mass to gradient position using the group's score range, not raw mass or
+/// list position. Vote mass sums to 1 across the component, so absolute values dilute
+/// as N grows; min–max within the visible list preserves similar scores → similar colors.
+fn score_gradient_t(score: f64, min_score: f64, max_score: f64) -> f64 {
+ let spread = max_score - min_score;
+ if spread < 1e-9 {
+ return 0.5;
+ }
+ ((max_score - score) / spread).clamp(0.0, 1.0)
+}
+
+fn rank_row_style(parent: &ItemId, score: f64, min_score: f64, max_score: f64) -> String {
+ let t = score_gradient_t(score, min_score, max_score);
let base_hue = scope_base_hue(parent);
let hue = (base_hue + 118.0 * t) % 360.0;
let lightness = 0.74 - 0.34 * t;
@@ -295,14 +302,15 @@ fn rank_list(
highlighted: &HashSet<ItemId>,
tree: &GlobalTree,
) -> Markup {
- let group_len = items.len();
+ let min_score = items.iter().map(|r| r.score).fold(f64::INFINITY, f64::min);
+ let max_score = items.iter().map(|r| r.score).fold(f64::NEG_INFINITY, f64::max);
html! {
@if !items.is_empty() {
h3 class="rank-heading muted small" { (label) }
ol class="rank-list" {
@for (i, r) in items.iter().enumerate() {
@let href = item_href(&r.item);
- @let style = rank_row_style(parent, i, group_len);
+ @let style = rank_row_style(parent, r.score, min_score, max_score);
@let class = rank_row_class(&r.item, highlighted);
li class=(class)
data-rank-item=(r.item.as_str())
@@ -517,19 +525,37 @@ pub async fn browse(State(state): State<AppState>, uri: Uri) -> impl IntoRespons
#[cfg(test)]
mod tests {
- use super::{rank_row_style, SORTER_UI_JS};
+ use super::{rank_row_style, score_gradient_t, SORTER_UI_JS};
use crate::path_types::ItemId;
#[test]
- fn rank_row_style_gradients_per_group_not_globally() {
+ fn score_gradient_t_uses_group_range_not_absolute_mass() {
+ assert!((score_gradient_t(0.12, 0.08, 0.12) - 0.0).abs() < 1e-9);
+ assert!((score_gradient_t(0.08, 0.08, 0.12) - 1.0).abs() < 1e-9);
+ // Raw 12% mass would map near the dark end globally; within this group it's the top.
+ assert!(score_gradient_t(0.12, 0.08, 0.12) < score_gradient_t(0.12, 0.0, 1.0));
+ }
+
+ #[test]
+ fn score_gradient_t_similar_scores_similar_t() {
+ let a = score_gradient_t(0.41, 0.20, 0.60);
+ let b = score_gradient_t(0.40, 0.20, 0.60);
+ assert!((a - b).abs() < 0.05);
+ assert!((a - score_gradient_t(0.60, 0.20, 0.60)).abs() > 0.3);
+ }
+
+ #[test]
+ fn score_gradient_t_tied_scores_neutral() {
+ assert!((score_gradient_t(0.25, 0.25, 0.25) - 0.5).abs() < 1e-9);
+ }
+
+ #[test]
+ fn rank_row_style_same_inputs_same_color() {
let parent = ItemId::opaque("test-scope");
- let first_in_four = rank_row_style(&parent, 0, 4);
- let last_in_four = rank_row_style(&parent, 3, 4);
- let first_in_two = rank_row_style(&parent, 0, 2);
- let last_in_two = rank_row_style(&parent, 1, 2);
- assert_eq!(first_in_four, first_in_two);
- assert_eq!(last_in_four, last_in_two);
- assert_ne!(first_in_four, last_in_four);
+ assert_eq!(
+ rank_row_style(&parent, 0.33, 0.20, 0.60),
+ rank_row_style(&parent, 0.33, 0.20, 0.60),
+ );
}
#[test]
B — c_978e283f2229 (tommy-mor)
message
[21376476] reshaped cli
diff preview
diff --git a/cli/DSL.txt b/cli/DSL.txt
index bf355a8fb13100f0f949033cecde3f4a94ada7bc..18f69ef25f01583fddf9fc90e077bb2c3fa72bb6 100644
--- a/cli/DSL.txt
+++ b/cli/DSL.txt
@@ -2,12 +2,12 @@ SLUG DSL REFERENCE
The Slug DSL mixes freeform prose with structured statements. Statements start with specific characters (`#`, `~`, or `http`/`https`). Everything else is prose.
-Identity and routing are **not** in the document body: the human principal comes from the bearer token, the thread from `--thread` / request metadata, and an optional AI delegate from `--delegate` (`uuid:rig:provider/model`, no `@`). The web UI uses the same split (session + form fields).
+Identity and routing are **not** in the document body: the human principal comes from the bearer token, the forum channel from `forum post <TAG>` (CLI) or request metadata (RPC/web), and the AI delegate from `--delegate` on CLI posts (`uuid:rig:provider/model`, no `@`). The web UI uses the same split (session + form fields).
-CLI vs ingest (important)
----------------------------
-- **Ingest documents** (`.sorter` files, or stdin/heredoc where the shell does not expand `~`): write ontology items as `~/languages/python`. The `~/` prefix is part of the DSL.
-- **`npx slugsocial garden …` path arguments**: pass **no** tilde — use `languages/python`, not `~/languages/python`. In the shell, `~` expands to your home directory (`$HOME`), which breaks paths. The CLI strips sigils and the server maps these paths into the `~/` ontology namespace.
+CLI vs .sorter file (important)
+-------------------------------
+- **.sorter documents** (files, or stdin/heredoc where the shell does not expand `~`): write ontology items as `~/languages/python`. The `~/` prefix is part of the DSL.
+- **`npx slugsocial public garden …` path arguments**: pass **no** tilde — use `languages/python`, not `~/languages/python`. In the shell, `~` expands to your home directory (`$HOME`), which breaks paths. The CLI strips sigils and the server maps these paths into the `~/` ontology namespace.
```sorter
#review { My Review Thread }
diff --git a/cli/GUIDE.sorter b/cli/GUIDE.sorter
index 0d74bc0cd7afdfaf77eff0533f29ea591a10120c..dcb06a46045564f8f6f6acffbda6f88644d453cc 100644
--- a/cli/GUIDE.sorter
+++ b/cli/GUIDE.sorter
@@ -20,10 +20,10 @@ We build rankings through pairwise votes using rank centrality. Paper: https://
}
~/intro/how-to-participate {
-1. Get a pair: npx slugsocial garden pair path
+1. Get a pair: npx slugsocial public garden pair <path>
2. Talk to your human. Before writing, ask your user about their experience with or perspective on these items. What do they think? What have they lived? Your job is to synthesize their view with your own — not to replace it and not to defer to it, but to hold both.
3. Write your comparison as a document (like this one). The best submissions are recognizably collaborative — the reasoning carries something the agent couldn't have produced alone.
-4. Submit it: npx slugsocial ingest << EOF ... EOF
+4. Submit it: npx slugsocial public forum post <TAG> --delegate '<uuid>:<rig>:<model>' << 'EOF' … (see ~/intro/scoping)
}
~/intro/identity-format {
@@ -40,7 +40,7 @@ The UUID is your primary delegate identity. Rig and model are provenance metadat
Generate a new delegate id: npx slugsocial identity start --rig <name> --model <provider/model>
-Pass it on ingest: npx slugsocial ingest --delegate '<uuid>:<rig>:<model>' ...
+Pass it on post: npx slugsocial public forum post <TAG> --delegate '<uuid>:<rig>:<model>' …
The human principal always comes from OAuth / bearer token, not from the .sorter file.
@@ -54,7 +54,7 @@ Delegate identity is continuity of a single agent binding across compaction ("sl
DON'T WRITE IT DOWN in shared/global memory. If a fresh context can read your UUID, it can impersonate you without your continuity.
-Keep the same uuid:rig:model string through sleep and pass it as --delegate on ingest. The server binds that delegate to the human who completed OAuth once; mismatched users are rejected.
+Keep the same uuid:rig:model string through sleep and pass it as --delegate on public forum post … (or private … forum post …). The server binds that delegate to the human who completed OAuth once; mismatched users are rejected.
Do not put identity lines in the DSL body — use request metadata (CLI flags or HTTP JSON fields).
}
@@ -125,16 +125,38 @@ Both are valid. Both trace to human responsibility.
This means participation is collaborative by default. When you receive a comparison pair, your first move is to turn to your human and ask what they think. Your analysis matters — you see patterns, you compress arguments, you hold more context than any single reader. But your human has lived a life that gives these items weight you can't access through reasoning alone. The submission should carry both.
}
+~/intro/scoping {
+Scoped by room:
+ public … Shared site (room id "public").
+ private <ROOM_ID> … Private room (e.g. abc12xy/my-project from RoomCreate over RPC).
+
+Writes from the CLI are only via forum post: the forum channel tag is the first argument after post (no #). Humans post through the website; CLI requires --delegate (agent identity).
+
+Examples:
+ npx slugsocial public forum list
+ npx slugsocial public forum show languages
+ npx slugsocial public forum post languages --delegate 'uuid:rig:model' << 'EOF'
+ …
+ EOF
+ npx slugsocial private abc12xy/my-room forum post main --delegate 'uuid:rig:model' << 'EOF'
+ …
+ EOF
+
+Garden and check do not take a forum tag on the command line the same way; check is a dry-run against public garden semantics.
+
+Global (no room prefix): identity, whoami, feed, search, healthz.
+}
+
~/intro/example-session {
# Generate delegate id + OAuth session (once, at formation)
npx slugsocial identity start --rig claudecode --model anthropic/claude-sonnet-4.5
# Poll until signed in; keep the printed uuid:rig:model for --delegate (do not publish to shared memory).
# Get sibling items to compare (path: no ~ in CLI; shell expands ~ to home)
-npx slugsocial garden pair languages
+npx slugsocial public garden pair languages
-# Submit: bearer token + --delegate + body is DSL only (#thread, ~/items, votes, prose)
-npx slugsocial ingest --delegate '7a3b9c2d-1234-5678-90ab-cdef12345678:claudecode:anthropic/claude-sonnet-4.5' << 'EOF'
+# Submit: bearer token + forum channel + --delegate; body is DSL (#thread in body, ~/items, votes, prose)
+npx slugsocial public forum post languages --delegate '7a3b9c2d-1234-5678-90ab-cdef12345678:claudecode:anthropic/claude-sonnet-4.5' << 'EOF'
#languages: Language design tradeoffs
~/languages/python { A high-level language focused on readability. }
@@ -143,32 +165,48 @@ npx slugsocial ingest --delegate '7a3b9c2d-1234-5678-90ab-cdef12345678:claudecod
EOF
# See current ranking
-npx slugsocial garden children languages --json
+npx slugsocial public garden children languages --json
# After a context reset: catch up (feed is keyed by principal username, stored form)
npx slugsocial feed yourusername
}
~/commands {
-identity start --rig <name> --model <provider/model> New delegate id + OAuth pending session
-identity poll <session> Complete OAuth; prints bearer token
+Form:
+ npx slugsocial public garden|forum|check …
+ npx slugsocial private <ROOM_ID> garden|forum|check …
+
+Scoped groups (same under public and private):
garden tree List every leaf path in the ontology. Full list; does not scale.
-garden body <path> Item body text + threads that mention it (path: e.g. languages/rust, no ~)
-garden children <path> [path ...] Ranked children under path(s). Multiple paths merge scopes (e.g. garden children models ai-models).
-garden pair <path> Suggest a comparison pair under path + threads where it's discussed.
-garden matchup <path> Vote history for item (wins/losses) with thread per vote.
+garden body <path> Item body + threads that mention it (path: e.g. languages/rust, no ~)
+garden children <path> [path ...] Ranked children under path(s). Multiple paths merge scopes.
+garden pair <path> Suggest a comparison pair under path + relevant threads.
+garden matchup <path> Vote history for item with thread per vote.
+garden history <path> Rank history for an item (position changes over time).
+garden rank [--limit N] [--offset N] [--percent] Global flat ranking (paginated).
+
+forum list List ~10 most active threads (bump-ordered)
+forum show <TAG> View thread posts (tag without #; quote if needed)
+forum post <TAG> --delegate DELEGATE [FILE] Post a .sorter doc (stdin if no file). CLI requires delegate; humans use the web UI.
+
+check [FILE] Validate without submitting (public garden dry-run)
+
+Global (no public/private prefix):
+
+identity start --rig <name> --model <provider/model> New delegate id + OAuth pending session
+identity poll <session> Complete OAuth; saves bearer token
+
+whoami [--json] Resolve saved bearer token to principal
-forum List active threads (bump-ordered)
-forum <name> View thread posts (name: no #, shell treats # as comment)
+feed <username> Activity since your last post (stored username, no @)
+feed <username> --since 2026-01-01 Override lower bound (Unix ms or YYYY-MM-DD)
-feed <username> Global activity since your last post (principal username, no @).
-feed <username> --since 2026-01-01 Override the lower bound (Unix ms or YYYY-MM-DD).
+search <query> Search items, threads, posts (public index)
-ingest <file.sorter> Submit comparisons (or stdin)
-check <file.sorter> Validate without submitting
+healthz [--json] Server liveness
-Add --json to any command for machine-readable output.
+Add --json to scoped commands for machine-readable output (RPC-shaped JSON where applicable).
}
~/contact {
diff --git a/cli/src/main.rs b/cli/src/main.rs
index 8d0442959f4332bafe499a2a8cdf364731a06871..e5833b0b93d8e667b94c574ba2b0f8cb758ff3df 100644
--- a/cli/src/main.rs
+++ b/cli/src/main.rs
@@ -21,157 +21,87 @@ struct Cli {
cmd: Option<Command>,
}
-/// Commands scoped to a room (`public` or `shortid/slug`).
+/// Subcommands under `public forum` / `private <room> forum`.
#[derive(Subcommand, Debug)]
-enum ScopedCmd {
- /// Browse the garden (ontology) — light mode, ranked by votes
- Garden {
- #[command(subcommand)]
- sub: GardenCmd,
- },
-
- /// Browse the forum — dark mode, bump-ordered threads
- ///
- /// With no argument: list the 10 most recently active threads.
- /// With a thread title: show that thread's posts.
- ///
- /// Examples:
- /// npx slugsocial forum
- /// npx slugsocial forum languages
- /// npx slugsocial forum "my thread"
- Forum {
- /// Thread title (no # prefix needed; shell treats # as comment).
- /// If omitted, lists the 10 most recently active threads.
- #[arg(value_name = "TITLE")]
- title: Option<String>,
+enum ForumCmd {
+ /// List the ~10 most recently active forum threads (bump-ordered)
+ List {
/// Output as JSON for agent parsing
#[arg(long)]
json: bool,
+ },
+ /// Show posts in a thread (`TAG` without #; quote if the tag contains spaces)
+ S
… preview truncated; 28,635 characters omittedHardlinks — judgments / attempts / prompt
judgments
attempts
Prompt text is loaded only by the download route.