PR #28618: feat(display): add tool_progress previews for fact_store and fact_feedback

• Repository: NousResearch/hermes-agent
• Author: zccyman
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/28618

Description (problem / solution / changelog)

Problem

When tool_progress is enabled on the Feishu platform, tools like fact_store and fact_feedback show unhelpful progress bubbles — only the tool name (e.g. ⚙️ fact_store...) with no indication of what action is being performed or what data is involved.

This is because build_tool_preview() in agent/display.py had no handler for these tools, returning None which falls through to a generic tool-name-only display.

Solution

Add action-based preview generation for both tools in build_tool_preview():

fact_store — 6 actions:

fact_feedback — shows action + fact ID:

• helpful: #136

Long content is truncated at 25-30 chars with ... suffix, matching the existing memory tool pattern.

Testing

• Added 12 new unit tests covering all fact_store actions + fact_feedback + truncation
• All 40 tests in test_display.py pass
• No security impact (pure display logic)

Fixes #28598

Changed files

• agent/display.py (modified, +24/-0)
• tests/agent/test_display.py (modified, +48/-0)

PR #28719: feat(display): declarative tool-preview schema to eliminate per-tool hardcoding (#28621)

• Repository: NousResearch/hermes-agent
• Author: xxxigm
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/28719

Description (problem / solution / changelog)

What does this PR do?

Adds a declarative tool-preview schema registry to agent/display.py so new tools — including plugin/third-party tools — can ship a one-line progress preview alongside their schema definition, instead of editing the hardcoded if-elif chain in build_tool_preview() every time.

Before, tools without a manual entry fell back to ⚙️ tool_name... on Feishu / Telegram bubbles and the CLI spinner. This is what prompted #28598 (fact_store / fact_feedback showing no useful context); #28621 generalises that observation: the if-elif chain is a structural oversight generator, and plugins have no path to preview support at all.

The new API is:

from agent.display import register_tool_preview

register_tool_preview(
    "fact_store",
    field="action",
    templates={
        "add":    '+ "{content:.30}"',
        "search": 'search: "{query:.25}"',
        "remove": "remove: #{fact_id}",
        "*":      "{action}",          # wildcard fallback
    },
    truncate=60,                        # per-tool cap
)

Design choices:

• Backward compatible — registry is consulted before the legacy chain. Existing tools keep working; migration is incremental, no flag-day.
• Plugin-friendly — plugins call register_tool_preview() at import time, next to their schema definition. No core code edit required.
• Safe by default — missing keys render as "" (a partial tool call never crashes the spinner), string values are whitespace-collapsed, lists are joined with , , bad templates degrade to None + a warning log instead of raising.
• Per-tool override — truncate=N takes priority over the global _tool_preview_max_len, useful for plugins that know their content is naturally short.
• Last-write-wins — plugins can override built-in previews if they need to.

Related Issue

Fixes #28621 Also closes #28598 (the user-visible bug that prompted this proposal — fact_store and fact_feedback now render meaningful previews).

Type of Change

• ✨ New feature (non-breaking change that adds functionality)
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)

Changes Made

• agent/display.py — adds register_tool_preview(), _unregister_tool_preview(), the _TOOL_PREVIEW_REGISTRY module-level dict, _SafePreviewArgs (a dict subclass that returns "" for missing keys and normalises values for str.format_map), and _render_registered_preview(). build_tool_preview() consults the registry first, applies the per-tool truncate (falling back to the global _tool_preview_max_len), and only then falls through to the existing hardcoded chain. Also hardens the legacy path against truthy non-dict args.
• plugins/memory/holographic/__init__.py — registers preview templates for fact_store (10 actions, including add, search, probe, reason, update, remove, plus a * fallback) and fact_feedback (helpful / unhelpful / *). The registration sits right next to FACT_STORE_SCHEMA / FACT_FEEDBACK_SCHEMA so reviewers see UX + schema together.
• tests/agent/test_display.py — adds TestRegisterToolPreview (15 unit tests covering single-template rendering, field dispatch with * fallback, missing-key safety, list-arg joining, per-field precision truncation, per-tool vs. global truncate precedence, whitespace collapsing, None → "", bad-template fallback with log assertion, last-write-wins re-registration, registry-over-legacy priority, and rejection of malformed registrations) plus TestFactStorePluginPreview (7 smoke tests for the holographic plugin registrations) and one extra defensive test for truthy non-dict args.
• website/docs/developer-guide/adding-tools.md — adds an "Optional: Tool Progress Previews" section showing the API and dispatch form, plus a checklist nudge so authors register a preview when relevant.

Memory and the other complex special cases (process, todo, send_message) are intentionally left on the legacy path in this PR — the issue's scope calls for 2–3 POC migrations with the rest deferred. memory in particular has a "<missing old_text>" sentinel that doesn't fit the minimal field/templates/truncate schema cleanly; if that's wanted, it should land in a follow-up PR (possibly extending the schema with an optional defaults mapping).

How to Test

# Full registry coverage + smoke tests for the plugin migration.
scripts/run_tests.sh tests/agent/test_display.py -q
# Expected: 55 passed (31 existing + 24 new).

# Manual: with the holographic-memory plugin enabled, watch a tool
# progress bubble while issuing a fact_store call from chat.
# Before this PR:  ⚙️ fact_store...
# After this PR:   ⚙️ fact_store + "user prefers tabs"

Checklist

Code

• My commit messages follow Conventional Commits (feat(display):, feat(memory/holographic):, test(display)+docs:)
• I searched for existing PRs — no duplicate
• My PR contains only changes related to this feature
• I've run scripts/run_tests.sh tests/agent/test_display.py -q (55 passed, 0 failed)
• I've added tests for my changes (22 new tests covering the registry contract + plugin smoke tests)
• I've tested on my platform: macOS 15.x (darwin 24.6.0)

Documentation & Housekeeping

• I've updated relevant documentation (website/docs/developer-guide/adding-tools.md)
• N/A — no config keys changed (cli-config.yaml.example)
• N/A — no architecture / workflow change requiring CONTRIBUTING.md / AGENTS.md updates
• I've considered cross-platform impact — pure Python, no platform-specific code paths
• N/A — no tool description/schema changes (preview registration is additive metadata)

Changed files

• agent/display.py (modified, +167/-0)
• plugins/memory/holographic/__init__.py (modified, +35/-0)
• tests/agent/test_display.py (modified, +209/-0)
• website/docs/developer-guide/adding-tools.md (modified, +46/-0)