saga-cli 0.1.0__tar.gz → 0.3.0__tar.gz

saga_cli-0.3.0/.github/CODEOWNERS

@@ -0,0 +1,3 @@
+# Every change requires review from the maintainer.
+# See https://docs.github.com/articles/about-code-owners
+* @JakeBeresford

{saga_cli-0.1.0 → saga_cli-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: saga-cli
-Version: 0.1.0
+Version: 0.3.0
 Summary: Generate a self-contained static HTML saga of a git diff — a chapter-by-chapter guided tour of a change.
 Project-URL: Homepage, https://github.com/JakeBeresford/saga
 Project-URL: Repository, https://github.com/JakeBeresford/saga
@@ -42,8 +42,11 @@ the data). Open it offline, email it, commit it, or drop it on any static host.
 
 - Python 3.11+
 - `git`
+- The [`gh`](https://cli.github.com) CLI, authenticated — only for reviewing a PR
+  by URL (`saga <pr-url>`) or pushing review comments
 - An API key for your chosen provider, in the standard environment variable:
-  `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, or `OPENROUTER_API_KEY`
+  `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, or `OPENROUTER_API_KEY` — or a running
+  local server (Ollama / LM Studio) and no key, with a `local/` model
 
 Generation is one structured LLM call made through [`instructor`](https://python.useinstructor.com).
 
@@ -64,7 +67,13 @@ e.g. `uv tool install /path/to/saga`. To upgrade: `uv tool install --force saga-
 
 ## Usage
 
-From inside the repo you want to review:
+From inside the repo, on the branch you want to review:
+
+```sh
+saga
+```
+
+From inside the repo, reviewing a different branch:
 
 ```sh
 saga --base main --head my-feature -o saga.html --open
@@ -76,14 +85,32 @@ Run it from anywhere with `--repo`:
 saga --repo ~/src/some-project --base main --head my-feature -o out.html
 ```
 
+### From a GitHub PR URL
+
+Point saga straight at a pull request by passing its URL as the first argument.
+No checkout is needed and it works from any directory:
+
+```sh
+saga https://github.com/owner/repo/pull/5
+saga https://github.com/owner/repo/pull/5 --model openai/gpt-4o -o pr5.html
+```
+
+This fetches the PR's diff, commits, and branch names with the
+[`gh`](https://cli.github.com) CLI (so `gh` must be installed and authenticated),
+then builds the saga exactly as it would for a local branch. In this mode the PR
+defines the change set, so `--base`, `--head`, and `--repo` are ignored.
+
+The optional positional argument is a GitHub PR URL (as above); all flags below
+apply to both modes.
+
 | Flag                   | Default                     | Meaning                                                                                            |
 | ---------------------- | --------------------------- | -------------------------------------------------------------------------------------------------- |
-| `--base`               | `main`                      | Base ref to diff against                                                                           |
-| `--head`               | current branch              | Head ref to walk through                                                                           |
+| `--base`               | auto-detected               | Base ref to diff against (defaults to the repo's default branch, e.g. `origin/main`); local mode only |
+| `--head`               | current branch              | Head ref to walk through; local mode only                                                          |
 | `--intent PATH`        | —                           | Optional plan/spec describing the change's intent, for plan-aware narration and deviation flagging |
 | `--model`              | `anthropic/claude-opus-4-8` | `provider/model` string (see [Providers](#providers)); also `$SAGA_MODEL`                          |
 | `-o, --output`         | `saga.html`                 | Output file                                                                                        |
-| `--repo`               | cwd                         | A path inside the target git repo                                                                  |
+| `--repo`               | cwd                         | A path inside the target git repo; local mode only                                                |
 | `--open` / `--no-open` | on                          | Open the result in a browser (on by default; `--no-open` to disable)                               |
 
 ## Providers
@@ -97,6 +124,8 @@ the matching API key:
 | Anthropic  | `anthropic/claude-opus-4-8`              | `ANTHROPIC_API_KEY`  |
 | OpenAI     | `openai/gpt-4o`                          | `OPENAI_API_KEY`     |
 | OpenRouter | `openrouter/anthropic/claude-3.5-sonnet` | `OPENROUTER_API_KEY` |
+| Local      | `local/qwen2.5-coder:14b`                | none                 |
+| Claude CLI | `claude-cli` or `claude-cli/sonnet`      | none (Claude Code login) |
 
 ```sh
 export SAGA_MODEL=openai/gpt-4o
@@ -104,6 +133,47 @@ export OPENAI_API_KEY=sk-…
 saga --base main --head my-feature -o saga.html
 ```
 
+### Local LLMs (Ollama / LM Studio)
+
+A `local/` model runs against any OpenAI-compatible local server, with no API
+key. Pull a capable coder model first (`ollama pull qwen2.5-coder:14b`), then:
+
+```sh
+saga --model local/qwen2.5-coder:14b --base main --head my-feature
+```
+
+`local/` defaults to Ollama's endpoint (`http://localhost:11434/v1`). Point it
+at another server — e.g. LM Studio — with `SAGA_LOCAL_BASE_URL`:
+
+```sh
+export SAGA_LOCAL_BASE_URL=http://localhost:1234/v1   # LM Studio
+saga --model local/your-loaded-model --base main --head my-feature
+```
+
+Two caveats: saga requires schema-valid JSON output, so use an instruction-tuned
+model that follows JSON prompting reliably; and the full diff plus a 16k output
+budget can exceed a small model's context window — prefer larger-context models
+and expect weaker narration than a frontier hosted model.
+
+### Claude Code CLI (no API key)
+
+If you don't have an Anthropic API key but you are logged into the
+[Claude Code](https://claude.com/claude-code) CLI — for example with a Claude
+Pro/Max subscription — the `claude-cli` model routes generation through
+`claude -p` instead of the API, reusing that login:
+
+```sh
+saga --model claude-cli --base main --head my-feature
+saga --model claude-cli/sonnet --base main --head my-feature   # pin a model
+```
+
+This shells out to the `claude` binary (which must be on your `PATH` and logged
+in), constrains its output to saga's schema via `--json-schema`, and runs it as
+a plain transform with no tools. `ANTHROPIC_API_KEY` and `ANTHROPIC_AUTH_TOKEN`
+are dropped for the subprocess so it uses your Claude Code login rather than
+silently billing an API key. It is slower than a direct API call (Claude Code
+boots an agent per run) and subject to your subscription's usage limits.
+
 ## As a Claude Code skill
 
 The `skills/` directory contains two Claude Code skills. To install both:
@@ -144,7 +214,8 @@ it on GitHub. Requires the [`gh`](https://cli.github.com) CLI, authenticated.
 
 ## How it works
 
-1. `diff.py` computes `git diff base...head` (no checkout) and the commit list.
+1. `diff.py` computes `git diff base...head` (no checkout) and the commit list —
+   or, given a PR URL, fetches the same diff and metadata from GitHub via `gh`.
 2. `model.py` splits the diff into stable-id hunks (`h0, h1, …`).
 3. `generate.py` sends the labeled diff + commits (+ optional intent) to the chosen
    model via `instructor`, which returns chapters as schema-validated JSON. Coverage
@@ -154,5 +225,4 @@ it on GitHub. Requires the [`gh`](https://cli.github.com) CLI, authenticated.
 
 ## Not included (yet)
 
-- Support for Local LLMs via ollama / LM Studio
 - A GitHub Action to auto-generate saga on PRs.

{saga_cli-0.1.0 → saga_cli-0.3.0}/README.md

@@ -15,8 +15,11 @@ the data). Open it offline, email it, commit it, or drop it on any static host.
 
 - Python 3.11+
 - `git`
+- The [`gh`](https://cli.github.com) CLI, authenticated — only for reviewing a PR
+  by URL (`saga <pr-url>`) or pushing review comments
 - An API key for your chosen provider, in the standard environment variable:
-  `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, or `OPENROUTER_API_KEY`
+  `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, or `OPENROUTER_API_KEY` — or a running
+  local server (Ollama / LM Studio) and no key, with a `local/` model
 
 Generation is one structured LLM call made through [`instructor`](https://python.useinstructor.com).
 
@@ -37,7 +40,13 @@ e.g. `uv tool install /path/to/saga`. To upgrade: `uv tool install --force saga-
 
 ## Usage
 
-From inside the repo you want to review:
+From inside the repo, on the branch you want to review:
+
+```sh
+saga
+```
+
+From inside the repo, reviewing a different branch:
 
 ```sh
 saga --base main --head my-feature -o saga.html --open
@@ -49,14 +58,32 @@ Run it from anywhere with `--repo`:
 saga --repo ~/src/some-project --base main --head my-feature -o out.html
 ```
 
+### From a GitHub PR URL
+
+Point saga straight at a pull request by passing its URL as the first argument.
+No checkout is needed and it works from any directory:
+
+```sh
+saga https://github.com/owner/repo/pull/5
+saga https://github.com/owner/repo/pull/5 --model openai/gpt-4o -o pr5.html
+```
+
+This fetches the PR's diff, commits, and branch names with the
+[`gh`](https://cli.github.com) CLI (so `gh` must be installed and authenticated),
+then builds the saga exactly as it would for a local branch. In this mode the PR
+defines the change set, so `--base`, `--head`, and `--repo` are ignored.
+
+The optional positional argument is a GitHub PR URL (as above); all flags below
+apply to both modes.
+
 | Flag                   | Default                     | Meaning                                                                                            |
 | ---------------------- | --------------------------- | -------------------------------------------------------------------------------------------------- |
-| `--base`               | `main`                      | Base ref to diff against                                                                           |
-| `--head`               | current branch              | Head ref to walk through                                                                           |
+| `--base`               | auto-detected               | Base ref to diff against (defaults to the repo's default branch, e.g. `origin/main`); local mode only |
+| `--head`               | current branch              | Head ref to walk through; local mode only                                                          |
 | `--intent PATH`        | —                           | Optional plan/spec describing the change's intent, for plan-aware narration and deviation flagging |
 | `--model`              | `anthropic/claude-opus-4-8` | `provider/model` string (see [Providers](#providers)); also `$SAGA_MODEL`                          |
 | `-o, --output`         | `saga.html`                 | Output file                                                                                        |
-| `--repo`               | cwd                         | A path inside the target git repo                                                                  |
+| `--repo`               | cwd                         | A path inside the target git repo; local mode only                                                |
 | `--open` / `--no-open` | on                          | Open the result in a browser (on by default; `--no-open` to disable)                               |
 
 ## Providers
@@ -70,6 +97,8 @@ the matching API key:
 | Anthropic  | `anthropic/claude-opus-4-8`              | `ANTHROPIC_API_KEY`  |
 | OpenAI     | `openai/gpt-4o`                          | `OPENAI_API_KEY`     |
 | OpenRouter | `openrouter/anthropic/claude-3.5-sonnet` | `OPENROUTER_API_KEY` |
+| Local      | `local/qwen2.5-coder:14b`                | none                 |
+| Claude CLI | `claude-cli` or `claude-cli/sonnet`      | none (Claude Code login) |
 
 ```sh
 export SAGA_MODEL=openai/gpt-4o
@@ -77,6 +106,47 @@ export OPENAI_API_KEY=sk-…
 saga --base main --head my-feature -o saga.html
 ```
 
+### Local LLMs (Ollama / LM Studio)
+
+A `local/` model runs against any OpenAI-compatible local server, with no API
+key. Pull a capable coder model first (`ollama pull qwen2.5-coder:14b`), then:
+
+```sh
+saga --model local/qwen2.5-coder:14b --base main --head my-feature
+```
+
+`local/` defaults to Ollama's endpoint (`http://localhost:11434/v1`). Point it
+at another server — e.g. LM Studio — with `SAGA_LOCAL_BASE_URL`:
+
+```sh
+export SAGA_LOCAL_BASE_URL=http://localhost:1234/v1   # LM Studio
+saga --model local/your-loaded-model --base main --head my-feature
+```
+
+Two caveats: saga requires schema-valid JSON output, so use an instruction-tuned
+model that follows JSON prompting reliably; and the full diff plus a 16k output
+budget can exceed a small model's context window — prefer larger-context models
+and expect weaker narration than a frontier hosted model.
+
+### Claude Code CLI (no API key)
+
+If you don't have an Anthropic API key but you are logged into the
+[Claude Code](https://claude.com/claude-code) CLI — for example with a Claude
+Pro/Max subscription — the `claude-cli` model routes generation through
+`claude -p` instead of the API, reusing that login:
+
+```sh
+saga --model claude-cli --base main --head my-feature
+saga --model claude-cli/sonnet --base main --head my-feature   # pin a model
+```
+
+This shells out to the `claude` binary (which must be on your `PATH` and logged
+in), constrains its output to saga's schema via `--json-schema`, and runs it as
+a plain transform with no tools. `ANTHROPIC_API_KEY` and `ANTHROPIC_AUTH_TOKEN`
+are dropped for the subprocess so it uses your Claude Code login rather than
+silently billing an API key. It is slower than a direct API call (Claude Code
+boots an agent per run) and subject to your subscription's usage limits.
+
 ## As a Claude Code skill
 
 The `skills/` directory contains two Claude Code skills. To install both:
@@ -117,7 +187,8 @@ it on GitHub. Requires the [`gh`](https://cli.github.com) CLI, authenticated.
 
 ## How it works
 
-1. `diff.py` computes `git diff base...head` (no checkout) and the commit list.
+1. `diff.py` computes `git diff base...head` (no checkout) and the commit list —
+   or, given a PR URL, fetches the same diff and metadata from GitHub via `gh`.
 2. `model.py` splits the diff into stable-id hunks (`h0, h1, …`).
 3. `generate.py` sends the labeled diff + commits (+ optional intent) to the chosen
    model via `instructor`, which returns chapters as schema-validated JSON. Coverage
@@ -127,5 +198,4 @@ it on GitHub. Requires the [`gh`](https://cli.github.com) CLI, authenticated.
 
 ## Not included (yet)
 
-- Support for Local LLMs via ollama / LM Studio
 - A GitHub Action to auto-generate saga on PRs.

{saga_cli-0.1.0 → saga_cli-0.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "saga-cli"
-version = "0.1.0"
+version = "0.3.0"
 description = "Generate a self-contained static HTML saga of a git diff — a chapter-by-chapter guided tour of a change."
 readme = "README.md"
 requires-python = ">=3.11"

{saga_cli-0.1.0 → saga_cli-0.3.0}/saga/assets/saga.css

@@ -13,16 +13,42 @@
 }
 .saga-theme-toggle:hover { border-color: var(--accent); color: var(--text); }
 
-.saga-crumbs { font-size: var(--fs-sm); color: var(--text-dim); margin-bottom: 10px; }
+.saga-crumbs { font-size: var(--fs-sm); color: var(--text-dim); margin-bottom: 8px; }
 .saga-crumbs a { color: var(--accent); text-decoration: none; }
 .saga-crumbs a:hover { color: var(--accent-bright); text-decoration: underline; }
-.saga-crumb-sep { margin: 0 8px; color: var(--text-faint); }
-.saga-crumb-current { color: var(--text); }
+
+.header h1 { line-height: 1.2; }
+
+.saga-summary {
+  color: var(--text-dim);
+  font-size: var(--fs-md);
+  line-height: 1.45;
+  margin-top: 4px;
+}
+
+.saga-statusline {
+  display: flex;
+  justify-content: space-between;
+  align-items: baseline;
+  gap: 8px 24px;
+  flex-wrap: wrap;
+  margin-top: 10px;
+}
 
 .saga-verdict {
   color: var(--text-dim);
   font-size: var(--fs-sm);
-  margin-top: 6px;
+  font-family: var(--font-mono);
+  letter-spacing: var(--tracking-mono);
+}
+.saga-verdict .saga-flag { color: var(--warn); }
+/* diff add/remove hues, matching the diff body's green/red */
+.saga-verdict .saga-add { color: var(--ok); }
+.saga-verdict .saga-del { color: var(--danger); }
+
+.saga-meta {
+  color: var(--text-faint);
+  font-size: var(--fs-xs);
   font-family: var(--font-mono);
   letter-spacing: var(--tracking-mono);
 }
@@ -86,15 +112,11 @@
   border: 1px solid var(--line); color: var(--text-dim); background: var(--surface-2);
 }
 .saga-plan { color: var(--text-dim); }
-.saga-qa { color: var(--ok); border-color: var(--ok-solid); background: var(--ok-tint); }
 .saga-read { color: var(--accent); border-color: var(--accent-solid); background: var(--accent-tint); }
-/* Deviations must be loud. */
-.saga-dev {
-  color: var(--danger-ink); background: var(--danger-loud); border-color: var(--danger); font-weight: 600;
-}
-/* Low confidence flagged for close review (amber). */
-.saga-low {
-  color: var(--warn-ink); background: var(--warn); border-color: var(--warn-solid); font-weight: 600;
+/* Agent attention flags (QA, differs-from-plan, closer-look) share one calm
+   amber; the label carries the meaning, not the colour. */
+.saga-qa, .saga-dev, .saga-low {
+  color: var(--warn); border-color: var(--warn-solid); background: var(--warn-tint);
 }
 
 /* --- chapter reader --- */
@@ -120,10 +142,10 @@
 .saga-chapter-head { display: flex; align-items: center; gap: 10px; flex-wrap: wrap; }
 .saga-chapter-title { font-size: var(--fs-xl); font-weight: 600; margin: 0; }
 
-/* Loud deviation banner. */
+/* Differs-from-plan banner (amber attention, same family as low-confidence). */
 .saga-deviation {
-  background: var(--danger-loud-bg); border: 1px solid var(--danger); border-left: 4px solid var(--danger);
-  color: var(--danger-ink); border-radius: var(--radius); padding: 12px 16px; margin: 14px 0; font-size: var(--fs-base);
+  background: var(--warn-tint); border: 1px solid var(--warn-solid); border-left: 4px solid var(--warn);
+  color: var(--warn); border-radius: var(--radius); padding: 12px 16px; margin: 14px 0; font-size: var(--fs-base);
 }
 .saga-lowconf {
   background: var(--warn-tint); border: 1px solid var(--warn-solid); border-left: 4px solid var(--warn);
@@ -139,7 +161,7 @@
   font-family: var(--font-mono); font-size: var(--fs-sm);
 }
 
-.saga-qanote { color: var(--ok); font-size: var(--fs-sm); margin: 8px 0; }
+.saga-qanote { color: var(--warn); font-size: var(--fs-sm); margin: 8px 0; }
 
 .saga-readmark {
   display: inline-flex; align-items: center; gap: 6px;

{saga_cli-0.1.0 → saga_cli-0.3.0}/saga/assets/saga.js

@@ -303,28 +303,65 @@
     chapters = data.chapters || [];
     comments = loadComments();
     initTheme();
-    renderVerdict(data.verdict);
+    renderHead();
+    renderVerdict(data.verdict, data.stats);
     renderTOC();
   }
 
+  // --- header (title, summary, provenance) ----------------------------
+
+  function renderHead() {
+    if (data.title) $('saga-title').textContent = data.title;
+    const sum = $('saga-summary');
+    if (sum && data.summary) { sum.textContent = data.summary; sum.hidden = false; }
+    const meta = $('saga-meta');
+    if (meta) {
+      const parts = [];
+      if (data.commit_sha) parts.push(data.commit_sha.slice(0, 7));
+      const rel = relativeTime(data.generated_at);
+      if (rel) parts.push('generated ' + rel);
+      meta.textContent = parts.join(' · ');
+    }
+  }
+
+  function relativeTime(iso) {
+    if (!iso) return '';
+    const then = new Date(iso).getTime();
+    if (isNaN(then)) return '';
+    const secs = Math.round((Date.now() - then) / 1000);
+    if (secs < 60) return 'just now';
+    const units = [['year', 31536000], ['month', 2592000], ['week', 604800],
+                   ['day', 86400], ['hour', 3600], ['minute', 60]];
+    for (const [name, size] of units) {
+      const n = Math.floor(secs / size);
+      if (n >= 1) return n + ' ' + name + (n === 1 ? '' : 's') + ' ago';
+    }
+    return 'just now';
+  }
+
   // --- verdict line --------------------------------------------------
 
-  function renderVerdict(v) {
+  function renderVerdict(v, stats) {
     if (!v) return;
-    const parts = [
-      v.chapters + (v.chapters === 1 ? ' chapter' : ' chapters'),
-      v.deviations + (v.deviations === 1 ? ' deviation' : ' deviations'),
-      v.low_confidence + ' low-confidence',
-    ];
-    if (v.qa && v.qa !== 'n/a') parts.push('QA ' + v.qa);
-    $('saga-verdict').textContent = parts.join(' · ');
-    // Shift the top status rail to the loudest state present:
-    // deviation (red) > attention (amber) > clear (green).
+    // Scope segments read neutral; attention flags echo the amber rail so a
+    // reviewer can spot what needs a look. All text is generated (no user
+    // content), so building innerHTML here is injection-safe.
+    const seg = [{ t: v.chapters + (v.chapters === 1 ? ' chapter' : ' chapters') }];
+    if (stats && stats.files) seg.push({ t: stats.files + (stats.files === 1 ? ' file' : ' files') });
+    if (stats && (stats.added || stats.removed)) {
+      seg.push({ html: '<span class="saga-add">+' + (stats.added || 0) + '</span> ' +
+                       '<span class="saga-del">−' + (stats.removed || 0) + '</span>' });
+    }
+    if (v.deviations > 0) seg.push({ t: v.deviations + (v.deviations === 1 ? ' differs from plan' : ' differ from plan'), flag: true });
+    if (v.low_confidence > 0) seg.push({ t: v.low_confidence + (v.low_confidence === 1 ? ' needs a closer look' : ' need a closer look'), flag: true });
+    $('saga-verdict').innerHTML = seg
+      .map((s) => (s.html ? s.html : s.flag ? '<span class="saga-flag">' + s.t + '</span>' : s.t))
+      .join(' · ');
+    // Two-tier status rail: amber when anything is flagged, else green.
     const rail = $('saga-rail');
     if (rail) {
-      rail.classList.remove('saga-rail-ok', 'saga-rail-attn', 'saga-rail-dev');
-      if (v.deviations > 0) rail.classList.add('saga-rail-dev');
-      else if (v.low_confidence > 0 || v.qa === 'attention') rail.classList.add('saga-rail-attn');
+      rail.classList.remove('saga-rail-ok', 'saga-rail-attn');
+      if (v.deviations > 0 || v.low_confidence > 0) rail.classList.add('saga-rail-attn');
       else rail.classList.add('saga-rail-ok');
     }
   }
@@ -334,9 +371,9 @@
   function badges(ch, read) {
     const out = [];
     if (ch.plan_step) out.push('<span class="saga-badge saga-plan">' + esc(ch.plan_step) + '</span>');
-    if (ch.deviation) out.push('<span class="saga-badge saga-dev">⚠ Deviation</span>');
-    if (ch.confidence === 'low') out.push('<span class="saga-badge saga-low">Low confidence</span>');
-    if (ch.qa && ch.qa.status === 'green') out.push('<span class="saga-badge saga-qa">✓ QA</span>');
+    if (ch.deviation) out.push('<span class="saga-badge saga-dev">Differs from plan</span>');
+    if (ch.confidence === 'low') out.push('<span class="saga-badge saga-low">Needs a closer look</span>');
+    if (ch.qa) out.push('<span class="saga-badge saga-qa">⚠ QA</span>');
     if (read) out.push('<span class="saga-badge saga-read">✓ Read</span>');
     return out.join('');
   }
@@ -390,13 +427,13 @@
     reader.hidden = false;
 
     const devBanner = ch.deviation
-      ? '<div class="saga-deviation"><strong>⚠ Deviation from plan.</strong> ' + esc(ch.deviation) + '</div>'
+      ? '<div class="saga-deviation"><strong>Differs from the plan.</strong> ' + esc(ch.deviation) + ' Worth confirming this was intentional.</div>'
       : '';
     const lowBanner = ch.confidence === 'low'
-      ? '<div class="saga-lowconf">Low confidence — this chapter needs close review.</div>'
+      ? '<div class="saga-lowconf"><strong>Needs a closer look.</strong> The walkthrough is unsure here — read the diff rather than just trusting the summary.</div>'
       : '';
-    const qaLine = ch.qa && ch.qa.note
-      ? '<div class="saga-qanote">' + (ch.qa.status === 'green' ? '✓ ' : '') + esc(ch.qa.note) + '</div>'
+    const qaLine = ch.qa
+      ? '<div class="saga-qanote">⚠ Manual QA: ' + esc(ch.qa) + '</div>'
       : '';
 
     // Class-based (not id-based) so the same nav can render at top and bottom.

{saga_cli-0.1.0 → saga_cli-0.3.0}/saga/assets/tokens.css

@@ -1,9 +1,9 @@
 /* Saga — design tokens ("instrument ink").
  *
  * Single source of truth for the saga page. A cool CYAN accent means
- * "a human can act here" (links, focus), and a strict three-light status system
- * — green (clear) · amber (attention) · red (deviation) — is reserved for the
- * agent's own state and never used as chrome.
+ * "a human can act here" (links, focus), and a two-light status system
+ * — green (clear) · amber (attention) — signals the agent's own state and is
+ * never used as chrome. Red is reserved for errors and destructive actions.
  *
  * Every colour is a light-dark(light, dark) pair. The page follows the OS by
  * default (color-scheme: light dark); the reader's toggle forces a scheme by
@@ -49,12 +49,11 @@
   --warn-ink:   light-dark(#ffffff, #1b1200);  /* text on a solid amber fill */
   --warn-tint:  light-dark(rgba(154,103,0,0.12), rgba(224,168,62,0.15));
 
-  --danger:        light-dark(#c62828, #f2685f);   /* deviation               */
+  --danger:        light-dark(#c62828, #f2685f);   /* errors, destructive     */
   --danger-bright: light-dark(#a51f1f, #ff9c94);
   --danger-ink:    light-dark(#5c1512, #ffe1de);   /* text on a loud red fill */
   --danger-tint:   light-dark(rgba(198,40,40,0.10), rgba(242,104,95,0.14));
-  --danger-loud:   light-dark(#fbe3e1, #b62324);   /* loud deviation fill     */
-  --danger-loud-bg:light-dark(#fdecea, #3d1518);   /* loud deviation panel bg */
+  --danger-loud-bg:light-dark(#fdecea, #3d1518);   /* error panel bg          */
   --danger-loud-border:light-dark(#c62828, #f2685f);
 
   /* --- diff shading (derived from status hues) --- */
@@ -129,7 +128,7 @@
 /* --- signature: the status rail ------------------------------------
  * A thin backlit edge across the top of the page. It shifts colour to
  * the most important state present:
- * accent (idle) · green (clear) · amber (attention) · red (deviation).
+ * accent (idle) · green (clear) · amber (attention).
  */
 .saga-rail {
   height: 2px;
@@ -138,4 +137,3 @@
 }
 .saga-rail-ok   { --rail: var(--ok); }
 .saga-rail-attn { --rail: var(--warn); }
-.saga-rail-dev  { --rail: var(--danger); }