feat: Rich Markdown rendering with skin-aware themes and /markdown toggle by lucaspirola · Pull Request #12836 · NousResearch/hermes-agent

lucaspirola · 2026-04-20T05:10:52Z

Summary

Adds full Rich Markdown rendering for CLI responses using Rich's Markdown class, with skin integration and user control.

Rich Markdown rendering — headings, bold, italic, code blocks (Pygments syntax highlighting), tables, lists, blockquotes rendered in the terminal
Skin-aware code themes — reads code_theme from the active skin; updates live on /skin change; falls back to monokai
Skin-aware text colour — reads banner_text as base paragraph colour so prose matches the theme while headings/code/bold keep their element styles
/markdown [on|off] command (alias /md) — toggle at runtime, persists to display.markdown in config.yaml
Fast-path plain-text detection — skips the parser when no markdown syntax is present (regex on first 500 chars)
Graceful fallback — all render calls wrapped in try/except; falls back to _RichText.from_ansi()
Streaming support — block-boundary detection avoids splitting mid-code-fence; fence state tracked across tokens
CLI platform hint — tells the LLM markdown is supported; adds cli_no_markdown variant for when /markdown off is active

Files changed (3)

File	Changes
`cli.py`	`_render_response()`, `_emit_stream_markdown()`, `_find_block_boundary()`, `_render_markdown_chunk()`, `_handle_markdown_command()`, skin theme init, `markdown_enabled` flag
`hermes_cli/commands.py`	`/markdown` + `/md` command registration
`agent/prompt_builder.py`	CLI and cli_no_markdown platform hint text

Supersedes #5150.

Test plan

Headings, bold, italic, code blocks, tables, lists render correctly
/markdown off shows raw syntax; /markdown on re-enables
/skin ares updates code theme colours live
Plain-text responses render unchanged (no parser overhead)
Config persists across sessions (display.markdown in config.yaml)

🤖 Generated with Claude Code

The CLI was displaying raw markdown syntax (**bold**, ```code```, etc.) because responses were rendered through _rich_text_from_ansi() which only interprets ANSI escapes. This adds full markdown rendering using Rich's built-in Markdown class with Pygments syntax highlighting. Changes: - Update CLI platform hint to encourage LLM markdown generation - Add _render_response() helper with fast-path regex detection - Render non-streaming responses via Rich Markdown in Panel - Add block-boundary streaming: accumulate tokens, render complete blocks through Rich Markdown, buffer incomplete trailing block - Add /markdown (alias /md) command to toggle on/off with persistence - Conditionally omit Panel style= when markdown enabled to preserve heading/code/bold colors from being washed out Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Make the markdown renderer adapt to the active skin's colour palette instead of hardcoding monokai/white. _render_response() now accepts code_theme and text_color from the skin — banner_text becomes the base paragraph colour (Rich Markdown's style= layers underneath element styles, preserving heading/code/bold colours), and code_theme falls back to monokai unless a skin overrides it via get_color("code_theme"). Zero changes to skin definitions or SkinConfig — existing skins and user-defined skins work automatically through get_color() fallbacks. Also adds docstrings and inline comments to all markdown rendering code (regex, fast-path, block boundary detection, chunk rendering, streaming strategy, flush, command handler) for clarity and maintainability. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The CLI now has a full Rich Markdown renderer, so the platform hint should tell the LLM to use markdown instead of discouraging it. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

…g, term width cache - Raise fast-path scan window from 500 to 8192 chars so agentic responses with plain-text preambles before tables/code blocks still get markdown rendering - Add cli_no_markdown platform hint; pass it when markdown_enabled=False so the LLM doesn't emit markdown that would display as raw syntax - Add logger.debug() to both renderer exception handlers so render failures leave a diagnostic trail - Cache terminal width at stream-open time to avoid a syscall per chunk - Clarify in-code comment: skin theme is captured per response, not mid-stream (/skin takes effect on the next response) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

alt-glitch · 2026-04-22T23:25:50Z

Multiple competing PRs for CLI markdown rendering exist (#5084, #5617, #1986) — needs consolidation decision.

alt-glitch · 2026-04-22T23:26:44Z

Multiple competing PRs for CLI markdown rendering exist (#5084, #5617, #1986) — needs consolidation decision.

alt-glitch · 2026-04-22T23:27:33Z

Multiple competing PRs exist.

teknium1

Thanks for the substantial CLI rendering work. Current main already has a tested Rich Markdown render mode via display.final_response_markdown (cli.py:2467-2502, tests/cli/test_cli_markdown_rendering.py:15-193), so this needs to be consolidated with that path rather than restoring the older renderer.

Problems

cli.py:2787-2801 persists fence state across complete rescans of the retained buffer. On the next token delta, the same unmatched opening fence toggles that state again; a blank line inside an unfinished fence can then be emitted as a completed block through cli.py:2888-2894.
cli.py:3206 selects cli_no_markdown only during agent creation, but /markdown changes only self.markdown_enabled (cli.py:6639-6644). Existing agents are reused, so the promised prompt change does not take effect in an already-started session.
No tests accompany the new renderer, streaming state machine, command, or prompt behavior.

Suggested changes

Re-scope onto current main's _render_final_assistant_content() / display.final_response_markdown implementation and preserve its ANSI, Windows-path, and table handling.
Rework fence detection to avoid rescanning retained text with accumulated state, then add token-boundary tests for incomplete backtick and tilde fences.
Keep any prompt-policy change deferred to a new session so the active conversation's cached system prompt remains stable.

Automated hermes-sweeper review.

teknium1 · 2026-07-12T15:19:18Z

+        in half — the block is held in buffer until the closing fence
+        arrives, then rendered in one piece with syntax highlighting.
+        """
+        in_fence = self._stream_md_fence_open


_stream_md_buf is retained and this function rescans it on every delta, but _stream_md_fence_open already reflects the previous full scan. An unmatched opening fence is therefore toggled twice on the next call, so a blank line inside an unfinished fenced block can become a false safe boundary. Derive fence state from the retained buffer on each scan, or scan only new input with a correctly maintained state.

teknium1 · 2026-07-12T15:19:18Z

-                platform="cli",
+                # Use a markdown-off platform hint when rendering is disabled
+                # so the LLM doesn't produce markdown that would display raw.
+                platform="cli" if self.markdown_enabled else "cli_no_markdown",


This prompt variant is selected only when AIAgent is created. /markdown off updates self.markdown_enabled but does not replace the existing agent, so the claimed no-Markdown instruction does not take effect after the first turn. Do not invalidate an active session's prompt cache; defer prompt-policy changes to a new session or keep this toggle display-only.

teknium1 · 2026-07-12T15:19:18Z

+    try:
+        # Use skin text colour as base style; "none" means default terminal colour
+        md_style = text_color if text_color else "none"
+        return _RichMarkdown(text, code_theme=code_theme, style=md_style)


This Markdown path passes the original text directly to Rich, unlike the current renderer, which normalizes ANSI through _rich_text_from_ansi(...).plain before constructing Markdown. Preserve that normalization and add an ANSI-bearing Markdown test.

lucaspirola and others added 4 commits April 20, 2026 07:02

feat: update CLI platform hint to encourage markdown formatting

cba846f

The CLI now has a full Rich Markdown renderer, so the platform hint should tell the LLM to use markdown instead of discouraging it. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

alt-glitch mentioned this pull request Apr 22, 2026

feat(cli): streaming markdown rendering in terminal output #1986

Closed

5 tasks

alt-glitch added type/feature New feature or request P2 Medium — degraded but workaround exists comp/cli CLI entry point, hermes_cli/, setup wizard labels Apr 22, 2026

This was referenced Apr 30, 2026

feat(cli): render markdown in responses with auto-detection #5980

Closed

feat: skin-aware code themes, /markdown toggle, fast-path & fallback (on top of KUSH42's renderer) #5084

Open

teknium1 reviewed Jul 12, 2026

View reviewed changes

doublerebel mentioned this pull request Jul 31, 2026

feat(cli): add width-safe streaming Markdown rendering #75326

Open

19 tasks

ericasw28 mentioned this pull request Aug 18, 2026

fix(cli): honor final_response_markdown: render while streaming #89081

Open

13 tasks

Navigation Menu

Uh oh!

feat: Rich Markdown rendering with skin-aware themes and /markdown toggle - #12836

feat: Rich Markdown rendering with skin-aware themes and /markdown toggle#12836
lucaspirola wants to merge 4 commits into
NousResearch:mainfrom
lucaspirola:feat/markdown-pr1-clean

lucaspirola commented Apr 20, 2026

Uh oh!

alt-glitch commented Apr 22, 2026

Uh oh!

alt-glitch commented Apr 22, 2026

Uh oh!

alt-glitch commented Apr 22, 2026

Uh oh!

teknium1 left a comment

Uh oh!

teknium1 Jul 12, 2026

Uh oh!

teknium1 Jul 12, 2026

Uh oh!

teknium1 Jul 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

lucaspirola commented Apr 20, 2026

Summary

Files changed (3)

Test plan

Uh oh!

alt-glitch commented Apr 22, 2026

Uh oh!

alt-glitch commented Apr 22, 2026

Uh oh!

alt-glitch commented Apr 22, 2026

Uh oh!

teknium1 left a comment

Choose a reason for hiding this comment

Problems

Suggested changes

Uh oh!

teknium1 Jul 12, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 Jul 12, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 Jul 12, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants