Dev to main, 13th August, 2026 - #808
Merged
Merged
Conversation
Adds the Create personas how-to under Simulation guides. It lays out the flow as annotated, step-by-step screenshots: open Personas and create one, choose Voice or Chat, fill the persona details and personality, tune the conversation settings, add custom properties, and find the new persona under the Custom tab. The page is scaffolded with Steps; step copy to follow. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds the Connect your agent how-to under Simulation guides. It lays out the flow as annotated, step-by-step screenshots: open Agent Definition and create one, choose the agent type and language, pick the provider, add the provider auth details, and set the behaviour prompt and commit message. The page is scaffolded with Steps; step copy to follow. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ighted Same fix as Agent definitions & versions: the locked spec wants one identical 3-piece diagram across all four Concept pages, this page had its own ad-hoc source-to-rows Mermaid instead.
Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Full jigsaw for the mental-model page, per the locked spec. Replaces the ad-hoc three-boxes-into-one Mermaid flowchart.
…nd fits sections Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
…xits Replace the scrapped jigsaw with the shared system diagram (agent in conversation with the FAGI simulator, fed by scenario and persona tables), saved as standalone SVGs with a highlighted copy per concept page. Scenarios drops the four build sources (how-to, now owned by the Create scenarios guide) and keeps definition, rows, simulator prompt, global nodes, and reuse. Both pages gain exits to the run/create/explore guides. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Grounded against origin/dev: a plain edit does not create a new version (only the explicit create-version action does), there is no compare-versions UI (only an internal AI-tool), and there is no restore/rollback button wired in the frontend. Reworded rollback as re-activating an older version, which is a real, UI-backed action.
Per the locked spec, all four pages (understanding-simulation, agent-definitions, scenarios, personas) need the identical 3-piece diagram with the relevant piece highlighted. This page had its own ad-hoc version-lifecycle Mermaid instead. Swapped it for the shared SVG jigsaw.
Ground the page in the product code and fix the reader-experience issues from the review passes: - drop version score/pass_rate/test_count (dormant model fields, no writer and no UI) and the Draft/Deprecated statuses (never set in the flow) - drop the websocket transport claim; websocket_url has no consumer in the run path - state the version lifecycle in order so "frozen" no longer contradicts editing the live configuration - weaken the rollback claim: activate() only flips status, it does not restore the definition's configuration - lead voice with the sufficient path (a phone number) before the native Vapi/Retell integrations - bullets for the definition contents and provider inputs, self-contained diagram caption, persona and SDK linked, versions and action exits wired Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Overview and Understanding Simulation now name the surface you drive Simulation from and route to the run guides instead of dead-ending in more concept pages. Scenarios defines the flow and its node types, so global nodes reads as the edge case it is, and states that a row pairs a persona with the case details. Agent definitions defines what an active version is and which one an unpinned run executes. Replay gains prerequisites, an exhaustive provider statement, and a circular loop diagram as a standalone SVG, replacing a Mermaid chain that restated the prose. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Ground the page in the product code and close the reader-experience gaps: - correct the simulator agent: it is not user-configurable, and the persona is the surface you shape it through - replace the bespoke Mermaid with the shared simulation model diagram, personas block highlighted - rewrite the worked example; the old one pinned fast speech, interrupting and price sensitivity onto an Indian accent - state how a persona reaches a run: scenarios pick personas at creation and carry their traits into the rows, so nothing is picked at launch - add the missing profession, languages and multilingual fields; background noise is background sound - "user" is a reserved term, so the simulated party is the customer - rename the Persona settings reference to Built-in personas, matching how Evaluation names its built-ins, and wire the reference and guide links Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Signed-off-by: khushalsonawat <khushalsonawat@gmail.com> # Conflicts: # src/lib/redirects.ts
Dataset, Prompt, Annotations, Knowledge Base, Agent Playground, Protect, Falcon AI and Optimization were rewritten into Concepts/Guides/Reference/ Troubleshooting but their sidebar groups still listed the old Features shape, so 91 new pages were on disk and unreachable. Replaces each group with the shape its manifest defines, including the nested explore-queue, build-workflow and reference/optimizers folders. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
…them Deletes the 48 pages the revamp replaces, adds a redirect for each one to a target verified to exist and be in the sidebar, and repoints inbound links across 75 files. Also repoints 49 legacy redirects that pointed at the deleted pages: /product/* and /future-agi/* URLs from the previous docs site now chain straight through to the new pages instead of landing on a 404. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
These pointed at paths that never existed in the new structure, several of them broken before this revamp: /docs/dataset/create, /docs/prompt/sdk, /docs/optimization/bayesian and /docs/quickstart among them. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Completes the sidebar locked in the implementation handoff: the four run and replay guides, evals in simulation, the three-page Explore results subfolder, Fix My Agent, both optimization guides, four reference pages and troubleshooting. Written from a verified fact-pack rather than from memory. The agent-simulate SDK surface was checked against the shipped PyPI release, not a repo branch, so the TestRunner and AgentInput signatures match what pip installs. LiveKit is absent throughout: it exists in the code but is not supported. Also renames the per-call tab to Call Details, or Chat Details on a chat run, across four pages that called it Logs. The canonical Explore results page and the frontend both use Call Details, so a reader following the old wording went looking for a tab that does not exist. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
…g at them Deletes the eight pages the new guides replace, adds a redirect for each to a target verified to exist, and repoints the three inbound links in faq.mdx, evaluation/guides/running-evaluations.mdx and get-started/connect-no-code-agents.mdx. Also chains eight legacy /product/simulation/* redirects that pointed at those now-deleted pages, so they land on the new guides instead of a 404. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Signed-off-by: khushalsonawat <khushalsonawat@gmail.com> # Conflicts: # src/lib/redirects.ts # src/pages/docs/simulation/concepts/scenarios.mdx
Applies the recorded review findings on the 19 remaining escalated pages across seven products, so every page in the revamp now has a final approve. Retires protect/concepts/guardrail-pipeline. Its diagram had pipeline stages and threshold-comparison edges, which is the internal mechanism the docs are not supposed to teach, and the threshold boundary it asserted was not in the verified facts. The vocabulary worth keeping moved out rather than being dropped: stage meanings and the threshold-vs-action distinction into understanding-protect, and the Fail Open / Mode / Timeout meanings into reference/guardrail-checks, which already owned their values and defaults. Five inbound links repointed, redirect added. Three pages the confirming review sent back, fixed against the product source rather than re-run: - create-skill: the original finding was wrong and had been applied, adding an Icon field to a form that has none. CustomizePanel renders four inputs; Icon belongs to SkillEditorDialog, a separate surface whose button reads Create. - optimizers/index: five default values disagreed with OptimizerConfigurationMapping, not the one that was flagged, and the table contradicted its own gepa.mdx sibling. - guardrail-checks: Provider Settings renders on any check carrying provider fields, so scoping it to provider-backed checks was wrong. Also corrects connect-mcp-server, which claimed discovery grants no permission to call a server's tools. Discovery enables every tool it finds, as the sibling concept page already said. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Simulation docs revamp
Bland.ai sits in the provider dropdown alongside Vapi and Retell, but the docs asserted in twelve places that only two providers exist. The backend SupportedProviders set was not used as evidence here, since it also lists LiveKit and ElevenLabs, which are not offered to customers. Records the three ways Bland differs: the Assistant ID field takes a Conversational Pathway ID, a contact number is always required because there is no web connector, and only one combined audio track is recorded, so an eval mapped to the stereo or per-channel recording resolves empty rather than failing loudly. Tool call evaluation stays Vapi-only. The Observe page is deliberately left at Vapi and Retell: Bland appears in ObservabilityProvider's choices, but so does LiveKit, and no user-facing picker was found to confirm it. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
The table carried thirteen columns and overflowed the content column, so scrolling right moved the Name column off screen and left every value unattributable to a persona. Splits it into three five-column tables, each keyed on Name: who they are, how they communicate, and voice behaviour. Also drops two leaks: Background noise rendered raw true/false, and the truncated Finished-speaking and Interrupt headers gave bare numbers with no indication they run from 1 to 10. Tables have no overflow container in global.css, unlike pre, so any wide table pushes the page body sideways. That is a separate site-wide fix. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
In-dataset annotation views are deprecated; annotation queues take dataset rows directly, offered as "From Datasets" in the add-items picker. Points the retired URL at the add-items guide, which documents that source, rather than at the annotations overview. Three legacy redirects already targeted this page. They are chained through to the new destination in the same change, so retiring it does not leave redirects aimed at a deleted page. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
The link audit fails on /docs/simulation/features/view-results and /docs/simulation/concepts/agent-definition. Both are redirect keys, so they resolve for a visitor, but an internal cross-link should reach the live page without a redirect hop. The second is also just the singular of a plural slug. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
Both Knowledge Base guides shipped without an asset. This adds the two videos TH-7517 calls for and the <video> slots to carry them. The recordings drove the flow with two .docx files under a knowledge base named "FAQ knowledge base", so the running example moves off the invented `support-policies` and its three files to match what the reader sees. The SDK guide follows the same name so all three guides share one example. Both videos are trimmed to cut the macOS file picker, which exposed a local Downloads listing, and the browser autofill dropdown over the Name field. Signed-off-by: khushalsonawat <khushalsonawat@gmail.com>
|
Automatic Review Skipped Too many files for automatic review. If you would still like a review, you can trigger one manually by commenting: |
hadarishav
approved these changes
Aug 13, 2026
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
3 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.
Pull Request
Description
Describe the changes in this pull request:
Checklist
Related Issues
Closes #<issue_number>