feat(sre-agent): Packaged deployment, Teams delivery, Autonomous mode, UAT-validated scheduled tasks

.copilot/skills/_archive/narrated-deck/SKILL.md

@@ -0,0 +1,122 @@
+---
+name: narrated-deck
+description: Consume per-slide YAML files (output of doc-to-yaml or release-deck) and produce a branded FinOps Toolkit PowerPoint deck plus TTS-ready transcript files. Use this skill whenever you have a directory of slide YAML files and want to turn them into a narrated training video deck, generate transcripts for downstream text-to-speech, build the PPTX from a YAML scene plan, or produce the final slide assets in the FinOps Toolkit visual style. Typically chained after doc-to-yaml, which produces the YAML from MS Learn docs. Outputs feed text-to-speech (per-scene .txt → .mp3), gpt-image-2 (optional alternate slide backgrounds), music, sound-effects, and ffmpeg for full video assembly.
+---
+
+# narrated-deck
+
+Consume a directory of per-slide YAML files and produce the narrated-video asset set.
+
+| Input | Output |
+|---|---|
+| `slides/*.yaml` (release-deck schema) | `slides.pptx` (FinOps Toolkit branded) |
+| | `transcripts/scene_NN_<slug>.txt` (TTS-ready, one per scene) |
+| | `scenes.json` (combined scene index for downstream skills) |
+
+This skill is the second half of the doc-to-video pipeline:
+
+```
+docs-mslearn/*.md
+   ↓ doc-to-yaml          (extract scenes, normalize voice, emit YAML)
+slides/*.yaml
+   ↓ narrated-deck        (this skill: build PPTX + write TTS transcripts)
+slides.pptx + transcripts/
+   ↓ text-to-speech       (each .txt → .mp3 via ElevenLabs / Azure)
+   ↓ gpt-image-2          (optional: replace slide backgrounds with brand art)
+   ↓ music + sound-effects (background bed + transitions)
+   ↓ ffmpeg               (compose PPTX-rendered PNGs + audio → MP4)
+final video
+```
+
+## When this skill triggers
+
+Any of:
+- "Build the deck from this YAML"
+- "Consume the YAML slides into a PPTX"
+- "Make the PowerPoint from `slides/`"
+- "Generate transcripts and PPTX from doc-to-yaml output"
+- "Render the narrated deck"
+
+## Usage
+
+```
+narrated-deck consume <slides_dir> --output <out_dir> [--next "Next module title"]
+```
+
+`<slides_dir>` is a directory containing `*.yaml` files in release-deck per-slide format. `<out_dir>` will receive `slides.pptx`, `transcripts/`, and `scenes.json`.
+
+## YAML schema (release-deck-compatible)
+
+Every input file has these keys:
+
+```yaml
+id: '1'
+order: 0
+cluster: overview
+asks_verbatim: []
+title: 'Heading text'
+content: '• Bullet 1<br/>• Bullet 2'
+notes: 'Spoken narration. Already normalized by upstream.<br/><br/>Second paragraph.'
+layout:
+  kind: BULLETS    # TITLE | BULLETS | TABLE | CODE | CALLOUT | OUTRO
+  addresses: []    # release-deck-only (MCAPS asks); ignored here
+  verdict: null    # release-deck-only; ignored here
+screens: null      # asset directives or null
+```
+
+The skill reads files in filename order. Each becomes one slide. Layout choice drives the renderer:
+
+| Layout | Renderer |
+|---|---|
+| `TITLE` | Hero slide with H1 (from `title`) + subtitle (first sentence of `notes`) |
+| `BULLETS` | Standard slide with `title` + `content` bullets |
+| `TABLE` | Standard slide with `title` + bullets (TABLE layouts work best when paired with hand-edited YAML carrying actual table data) |
+| `CODE` | Standard slide with `title` + monospace block |
+| `CALLOUT` | Section header with pull-quote |
+| `OUTRO` | Closing slide (uses `--next` flag for "Coming up next" copy) |
+
+Speaker notes pane is populated from the `notes` field verbatim — assumes upstream (doc-to-yaml) already applied the Microsoft spoken-voice rules.
+
+## Brand identity
+
+Slides inherit the FinOps Toolkit master template bundled at `assets/source-template-clean.pptx`:
+- Purple gradient (`#7B33C0` → `#4E2E7A`)
+- FinOps Toolkit hexagon logo
+- Microsoft Learn typography
+
+The clean template is generated from the original release-deck `source-template.pptx` by `scripts/prepare_template.py`, which strips orphan slide parts that would otherwise corrupt the output PPTX (causing PowerPoint's "found a problem with content" repair prompt).
+
+## Output layout
+
+```
+<out_dir>/
+├── slides.pptx                    ← branded editable deck
+├── scenes.json                    ← combined scene index
+└── transcripts/
+    ├── scene_01_intro.txt         ← TTS-ready, with <break time="1300ms" /> tags
+    ├── scene_02_what_you_get.txt  ← (slide-only marker if narration_only=true)
+    └── ...
+```
+
+## Transcripts
+
+One file per scene at `transcripts/scene_NN_<slug>.txt`. Format:
+
+- Prose-bearing scenes: the YAML `notes` field with `<break time="1300ms" />` between paragraphs
+- Slide-only scenes (table/code-dominated, no narration): a marker file starting with `# slide-only`
+
+The break tag is SSML — interpreted natively by ElevenLabs and Azure Speech Service. For consumers that don't speak SSML, run `transcript.py --no-breaks` to omit the tags.
+
+## Constraints
+
+- **No paraphrasing.** Speaker notes are written verbatim from the YAML. Voice rules are doc-to-yaml's responsibility, not this skill's.
+- **No source-doc access.** This skill only reads YAML; it doesn't touch the original `.md`.
+- **Brand-template-bound.** Output uses the bundled `assets/source-template-clean.pptx`. To use a different brand template, replace that file (or run `prepare_template.py` on a new master).
+
+## Constraints on YAML input
+
+- File extension must be `.yaml` (not `.yml`)
+- `layout.kind` must be one of: `TITLE`, `BULLETS`, `TABLE`, `CODE`, `CALLOUT`, `OUTRO`
+- `notes` should be in spoken voice (anaphora, articles, no source citations) — this skill does NOT re-normalize
+- For `TABLE` and `CODE` slides, the visual will currently render as a bullet list. Hand-edit the YAML or use release-deck/build_yaml.py for richer rendering of those layouts.
+

.copilot/skills/_archive/narrated-deck/evals/README.md

@@ -0,0 +1,52 @@
+# narrated-deck — evals
+
+This directory contains test cases and reference inputs for the `narrated-deck` skill.
+
+## Test inputs
+
+`inputs/` contains three SRE agent docs of varying complexity, used as fixtures across the eval set:
+
+| Doc | Size | Why it's in the eval set |
+|---|---|---|
+| `overview.md` | ~1.1k words, 5 scenes | Smallest — exercises the basic pipeline + table-driven scene detection |
+| `get-started.md` | ~1.1k words, 7 scenes | Medium — exercises the normalizer (run-on splitting on prose with semicolons) |
+| `kusto-tools.md` | ~2.7k words, 25 scenes | Largest — stress-tests the per-doc subagent throughput and confirms source-citation parenthetical removal at scale |
+
+These fixtures are snapshots of `docs-mslearn/toolkit/sre-agent/*.md` taken after the Microsoft Style Guide cleanup commit `ed295c3d`. They're committed alongside the skill so eval results stay reproducible even if the source docs evolve.
+
+## Running the evals
+
+The recommended path is per-doc subagent fan-out — spawn one subagent per eval, each invoking the skill on its assigned input. The orchestrator collates results.
+
+For a quick local check (single doc, no subagents):
+
+```bash
+python3 .copilot/skills/narrated-deck/scripts/convert.py \
+    .copilot/skills/narrated-deck/evals/inputs/overview.md \
+    --output .copilot/skills/narrated-deck/evals/outputs/eval-1
+```
+
+For the full batch (still single-process — subagent fan-out is the calling agent's responsibility):
+
+```bash
+python3 .copilot/skills/narrated-deck/scripts/convert.py \
+    .copilot/skills/narrated-deck/evals/inputs/overview.md \
+    .copilot/skills/narrated-deck/evals/inputs/get-started.md \
+    .copilot/skills/narrated-deck/evals/inputs/kusto-tools.md \
+    --output .copilot/skills/narrated-deck/evals/outputs/batch
+```
+
+## What to verify in the outputs
+
+For each eval directory, check:
+
+1. **`scenes.json`** — scene count matches the doc's H2 count (minus dropped sections). `narration_only: true` flags appear on table/code-dominated scenes.
+2. **`transcripts/scene_NN_*.txt`** — prose scenes carry natural-sounding narration with `<break time="1300ms" />` between paragraphs. Slide-only scenes are marker files starting with `# slide-only`.
+3. **`deck-outline.md`** — 7-cell table parses cleanly (no broken pipes from unescaped `|` in cell content). Speaker notes reproduce the transcript content with `<br/>` instead of `<break>`.
+4. **`slides.pptx`** — opens in PowerPoint. First slide is the title. Last is the outro. Each H2 from the source is a slide. Speaker notes pane shows the narration.
+
+## Expected qualitative observations
+
+- `overview.md` → 5 prose-bearing scenes + 1 slide-only ("What you get") + outro
+- `get-started.md` → 6 prose-bearing scenes + 1 slide-only ("Automation map") + outro
+- `kusto-tools.md` → ~25 scenes; "Source validation" is table-only; each tool subsection is a separate `BULLETS` scene

.copilot/skills/_archive/narrated-deck/evals/evals.json

@@ -0,0 +1,26 @@
+{
+  "skill_name": "narrated-deck",
+  "evals": [
+    {
+      "id": 1,
+      "name": "overview-short-doc-with-table-and-architecture-flow",
+      "prompt": "Convert the SRE agent overview doc at .copilot/skills/narrated-deck/evals/inputs/overview.md into TTS-ready transcripts and a slide deck. Output goes to .copilot/skills/narrated-deck/evals/outputs/eval-1/. Apply the canonical FinOps Toolkit voice rules.",
+      "files": [".copilot/skills/narrated-deck/evals/inputs/overview.md"],
+      "expected_output": "Per-doc output dir with scenes.json (5 scenes), deck-outline.md (7-cell table), slides.pptx (5+ slides + outro), and transcripts/scene_NN_*.txt (one per scene; 'What you get' should be a slide-only marker file because that section is table-dominated)."
+    },
+    {
+      "id": 2,
+      "name": "get-started-medium-doc-exercises-normalizer",
+      "prompt": "Convert the SRE agent get-started doc at .copilot/skills/narrated-deck/evals/inputs/get-started.md into TTS-ready transcripts and a slide deck. Output goes to .copilot/skills/narrated-deck/evals/outputs/eval-2/. Make sure the normalizer splits semicolon run-ons in the prose.",
+      "files": [".copilot/skills/narrated-deck/evals/inputs/get-started.md"],
+      "expected_output": "Output dir with all four artifacts. Transcripts must contain natural anaphora ('the agent' / 'It' on second mention). 'Automation map' section is table-dominated and should produce a slide-only marker."
+    },
+    {
+      "id": 3,
+      "name": "kusto-tools-long-reference-doc-stress-test",
+      "prompt": "Convert the SRE agent kusto-tools doc at .copilot/skills/narrated-deck/evals/inputs/kusto-tools.md into TTS-ready transcripts and a slide deck. Output goes to .copilot/skills/narrated-deck/evals/outputs/eval-3/. This is a 21-tool reference doc — the slide deck should have one slide per tool subsection (so ~25 slides total).",
+      "files": [".copilot/skills/narrated-deck/evals/inputs/kusto-tools.md"],
+      "expected_output": "Output dir with all four artifacts. ~25 scenes in scenes.json. The deck handles tools as BULLETS scenes with the 'Source YAML:' link preserved as the only inline source reference per tool. No '([source YAML](path))' parentheticals appear in any transcript."
+    }
+  ]
+}