hermes - 💡(How to fix) Fix [Skills System] Serious systemic defects in the skills system: no pruning, island isolation, and missing conflict detection [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
NousResearch/hermes-agent#12877Fetched 2026-04-20 12:16:27
View on GitHub
Comments
0
Participants
1
Timeline
0
Reactions
0

Root Cause

| Problem | Root Cause | Consequence | |---------|-----------|-------------| | Too low barrier | 5 calls = created | Invalid skills flood | | No decay mechanism | Only add, never clean | System bloat, heat death | | Skill islands | No composition/sharing | Duplicate code, hard to maintain | | Missing conflict detection | Only checks same name, not same function | Chaotic output, internal friction |

RAW_BUFFERClick to expand / collapse

markdown ## Source

 A user on Toutiao (Chinese social platform) named "小胖有事说" published a video titled "Hermes seems to be truly half-baked!" (2026-04-20),
 providing concentrated feedback on design flaws in Hermes's skill system. The original video has gained some attention and deserves official
 acknowledgment.

 ---

 ## 1. Unreasonable skill creation mechanism: "Only births, no nurturing"

 - **Too low barrier, proliferation of throwaway skills**: As few as 5 tool calls automatically convert temporary experience into a skill. One-off
 debugging sessions or accidentally successful tasks can all trigger this, leading to an overwhelming flood of invalid, temporary skills that bury
 the truly useful ones.
 - **Experience ≠ Skill, lacks consolidation and validation**: Short-term working memory is directly converted to long-term skill with no memory
 consolidation, effectiveness verification, or quality filtering — highly unscientific.

 **Recommendation**: Add quality gates for skill creation, such as requiring N successful uses before creation, placing newly created skills in a
 pending review state, or distinguishing "temporary skills" from "正式技能 (formal skills)".

 ---

 ## 2. No metabolism: only adding, never subtracting — system bloat

 - **No deletion/disable/invalidation mechanism**: Skills created once live forever; users can only delete manually and have no way to know which
 ones to remove.
 - **Lacks regulatory feedback loops**: Violates principles of system stability (cf. "Thinking in Systems"). Nature relies on "death" cycles, but
 Hermes only has additions with no cleanup. Eventually the skill library explodes, making the system slower and dumber.

 **Recommendation**: Introduce skill decay/invalidation (e.g., skills unused for 30 days are auto-demoted or flagged for cleanup), support
 disabling skills, and add usage statistics for user reference.

 ---

 ## 3. Skill islands: no composition, no sharing, redundant duplication

 - **No composition/inheritance/synergy**: Each skill is an isolated island — cannot call each other, share code, or logic.
 - **Massive redundant copying**: The same logic (e.g., TDD, development workflows) is duplicated across multiple skills; updating one doesn't sync
 to other copies.
 - **Official association fields are non-functional**: Fields like `related_skills` exist only for display, with no actual code-level invocation —
 purely decorative.

 **Recommendation**: Support inter-skill dependencies/calls, support shared code libraries (like npm packages), and `related_skills` fields should
 actually trigger cross-skill collaboration or at least provide intelligent recommendations.

 ---

 ## 4. Missing conflict detection: mutual interference, severe internal friction

 - **Only checks name conflicts, not functional conflicts**: Only skills with identical names are blocked. Skills with nearly identical
 functionality but different names are completely undetected.
 - **Skills conflict with and override each other**: Multiple drawing/task skills conflict and vie for execution rights. Some even have `conflict
 with` markers in the source code, resulting in chaotic outputs (e.g., image generation styles are completely inconsistent).

 **Recommendation**: Add functional similarity detection (based on semantic similarity of skill descriptions/tags), explicitly declare skill mutual
 exclusivity, and runtime conflict detection should give users clear selection prompts.

 ---

 ## 5. Overall Consequences

 The skill ecosystem is imbalanced, polluted, and riddled with internal friction. The system grows more bloated, more foolish, and less stable with
 use. The automated skill creation, meant to be a highlight, has become the biggest hidden danger due to lack of maintenance, cleanup, and
 coordination.

 ---

 ## Summary

 | Problem | Root Cause | Consequence |
 |---------|-----------|-------------|
 | Too low barrier | 5 calls = created | Invalid skills flood |
 | No decay mechanism | Only add, never clean | System bloat, heat death |
 | Skill islands | No composition/sharing | Duplicate code, hard to maintain |
 | Missing conflict detection | Only checks same name, not same function | Chaotic output, internal friction |

 Recommended the official treat "skill lifecycle management" as a dedicated project, redesigning the skill ecosystem with reference to the three
 essentials of systems (elements, connections, function/goal) from "Thinking in Systems."

 ---

 *Feedback source: https://www.toutiao.com/video/7630643722442654248/*

extent analysis

TL;DR

Implement a skill lifecycle management system with quality gates, decay mechanisms, and conflict detection to address the issues in the Hermes skill ecosystem.

Guidance

  • Introduce quality gates for skill creation, such as requiring a minimum number of successful uses before creation, to prevent the proliferation of invalid skills.
  • Implement a decay mechanism, such as auto-demoting or flagging skills for cleanup after a certain period of inactivity, to prevent system bloat.
  • Develop a conflict detection system that checks for functional similarities between skills, not just name conflicts, to prevent internal friction and chaotic outputs.
  • Consider redesigning the skill ecosystem to support inter-skill dependencies, shared code libraries, and intelligent recommendations based on related skills.

Example

No specific code snippet can be provided without more context, but an example of a quality gate could be:

def create_skill(experience):
    if experience.successful_uses < 5:
        return "Insufficient experience to create a skill"
    # Create skill logic here

Notes

The provided recommendations are based on the issues described in the feedback and may require significant changes to the existing system. It is essential to carefully evaluate and test any changes to ensure they do not introduce new issues or negatively impact the user experience.

Recommendation

Apply a workaround by introducing a basic skill lifecycle management system, focusing on quality gates and decay mechanisms, to mitigate the most critical issues. This will allow for a more stable and maintainable skill ecosystem while further development and redesign of the system can be planned and implemented.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

hermes - 💡(How to fix) Fix [Skills System] Serious systemic defects in the skills system: no pruning, island isolation, and missing conflict detection [1 participants]