Skill index size reduction and retrieval strategies
NousResearch/hermes-agent · cluster 2045 · agent
Data as of
13 open items. Fix attempted, not merged.
The short answer
9open issues
4open PRs
23items grouped
- Known?
- Yes. 23 related items grouped here, 1 issue already closed.
- Fix event
- No fix has been merged to main. 9 PRs closed without merging; see the attempts below.
- Tracker
- The maintainers' tracker marks this
unresolved. That is separate from the fix event. - Still open
- 9 issues and 4 PRs.
Next step: read the attempts below. Your case may extend or revive that work.
Pick up a Lab taskThese issues and PRs explore various approaches to reducing token overhead from skill listings in system prompts, including lazy loading, category-level indexing, semantic/FTS5 retrieval, skill graphs, and options to suppress skill injection entirely.
Open issues9
- Issue#2045[Feature]: Lazy skill loading: remove skill listing from system prompt, use on-demand tool instead
- Issue#17649Feature: Semantic Skill Retrieval with SQLite FTS5 — Replace 4500-token broadcast with on-demand search
- Issue#22620Feature Request: Skill list bloat causes massive context window inflation — need vector-based skill routing or lazy loading
- Issue#26786perf: two-level skill index in system prompt to reduce fixed token overhead
- Issue#26806[Feature] Suppress <available_skills> system_prompt injection for scripted/oneshot use
- Issue#34823Feature Request: Semantic / Per-Message Skill Retrieval
- Issue#37227[Feature]: Category-Aware Smart Skill Indexing with Automatic Scenario Matching + Lazy Loading (fusing #3879's trigger keywords)
- Issue#49967[Feature]: Skill Graph — optional dynamic skill discovery mode to reduce system prompt size
- Issue#63153[Feature]: Add Dynamic Skill Retrieval System (Skill-RAG) for Hermes Agent
Open PRs4
9 PRs closed without merging
From the close comments: won't fix / by design 2 · done elsewhere 1 · stale or abandoned 1 · superseded by another PR 1 · other 4
- #2089 feat(skills): add lazy skill loading to reduce system prompt tokens · other
-
#2420 feat: docker_extra_args, message timestamps, and lazy skill loading · other →
ebf2ea5,#23599 - #18316 feat: semantic skill retrieval with FTS5 + hybrid selector + skill enforcement · done elsewhere
- #26815 feat(agent): add --no-skills-index to suppress <available_skills> system-prompt block · stale or abandoned
-
#26842 perf(skills): list categories in system prompt · other →
ee1a744,#72813
Show 4 more
-
#28929 fix(anthropic): default skills catalog to names-only to reduce prompt token usage · superseded by another PR →
f64e4f4 -
#35345 feat(skills): add category-level skill index to reduce per-turn token overhead by 90% · other →
ee1a744,#26842 - #83314 feat(skills): add configurable skill index injection · won't fix / by design
- #130118 feat: skill-rag plugin — dynamic skill retrieval via embeddings (implementation for #63153) · won't fix / by design
Closed issues · 1
- Closed#28902OAuth Anthropic Max: <available_skills> system-prompt injection on every turn triggers "out of extra usage" 400 when skills/session_search toolsets are active
How this cluster was grouped
-
Link evidence (
partly structural links): some members are connected by structural evidence (shared changed files, or a closing/superseded reference); the rest are joined by an automated relevance judgement over their text. - Members outside that structural core may belong to a different underlying problem.
-
Pairwise link density
0.07: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36.1.00means every pair is linked.
Pick up a Lab task
Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.
Open the LabAt a glance
- Subsystem
- agent
- Closed issues
- 1
- Attempts
- 9 PRs closed without merging