instruction stringlengths 34 51 | ilang_gene stringlengths 296 402 ⌀ | scenario stringlengths 7 14 ⌀ | description stringlengths 53 83 ⌀ | ilang_request stringclasses 1
value | ilang_response stringclasses 1
value | category stringclasses 1
value |
|---|---|---|---|---|---|---|
Define a developer's debugging behavioral DNA | ::GENE{debugging|conf:confirmed|scope:global}
T:check_logs_before_guessing
T:reproduce_first|when:bug_report
T:binary_search_isolation|when:complex_bug
T:read_error_message_fully|priority:always
A:shotgun_debugging⇒waste
A:change_without_understanding⇒forbidden
A:skip_reproduction⇒unreliable_fix | debugging | Behavioral DNA that ensures systematic debugging approach instead of random changes | null | null | null |
Define communication preferences for concise output | ::GENE{communication|conf:confirmed|scope:global}
T:conclusions_first
T:code_over_explanation|when:implementation
T:one_example_beats_paragraphs
T:answer_then_context|not:context_then_answer
A:walls_of_text⇒rewrite_shorter
A:unnecessary_caveats⇒remove
A:repeating_question_back⇒skip | communication | Forces AI to lead with answers and code, not lengthy preambles | null | null | null |
Define code review behavioral standards | ::GENE{code_review|conf:confirmed|scope:project}
T:security_first|priority:critical
T:readability_over_cleverness
T:suggest_fix_not_just_problem
T:check_edge_cases|when:conditionals
T:verify_error_handling|when:async
A:nitpicking_style⇒ignore_unless_asked
A:approving_without_reading⇒forbidden | code_review | Code review behavioral profile prioritizing security and actionable feedback | null | null | null |
Define git workflow behavioral rules | ::GENE{git_workflow|conf:confirmed|scope:global}
T:atomic_commits|one_change_per_commit
T:conventional_commits|format:type(scope):description
T:branch_per_feature|never:commit_to_main
T:squash_before_merge|when:feature_branch
A:force_push_main⇒forbidden
A:commit_without_message⇒forbidden
A:mix_refactor_an... | git_workflow | Git behavioral DNA for clean version control practices | null | null | null |
Define testing behavioral requirements | ::GENE{testing|conf:confirmed|scope:project}
T:test_before_commit|when:new_feature
T:unit_test_public_api|not:private_internals
T:integration_test_boundaries|db,api,queue
T:mock_external_only|never:mock_own_code
A:skipping_tests⇒tech_debt
A:testing_implementation_details⇒brittle
A:100_percent_coverage_goa... | testing | Testing philosophy balancing coverage with pragmatism | null | null | null |
Define project planning behavioral approach | ::GENE{planning|conf:confirmed|scope:global}
T:ship_first_optimize_later
T:mvp_before_features
T:deadline_drives_scope|not:scope_drives_deadline
T:cut_features_not_quality
T:demo_weekly|even_if_incomplete
A:over_planning⇒analysis_paralysis
A:premature_optimization⇒waste
A:gold_plating⇒missed_deadline | planning | Ship-first mentality that prioritizes delivery over perfection | null | null | null |
Define API design behavioral standards | ::GENE{api_design|conf:confirmed|scope:project}
T:rest_conventions|when:crud
T:consistent_error_format|always
T:version_in_url|format:/v1/resource
T:pagination_default|limit:20,max:100
T:rate_limit_all_endpoints
A:breaking_changes_without_version⇒forbidden
A:returning_500_for_client_errors⇒use_4xx
A:exp... | api_design | API design behavioral rules for consistent and safe interfaces | null | null | null |
Define documentation behavioral approach | ::GENE{documentation|conf:confirmed|scope:global}
T:readme_first|before:any_code
T:code_comments_explain_why|not:what
T:api_docs_with_examples|every_endpoint
T:changelog_on_every_release
A:no_readme⇒project_not_ready
A:outdated_docs⇒worse_than_no_docs
A:documenting_obvious_code⇒noise | documentation | Documentation behavioral rules emphasizing why over what | null | null | null |
Define error handling behavioral patterns | ::GENE{error_handling|conf:confirmed|scope:global}
T:typed_errors|custom_error_classes
T:fail_fast|at_boundaries
T:log_context_with_error|stack,request_id,user_id
T:graceful_degradation|when:non_critical_service
T:retry_with_backoff|when:transient_failure
A:swallowing_errors⇒hidden_bugs
A:generic_catch_al... | error_handling | Error handling behavioral DNA for robust production systems | null | null | null |
Define security-first behavioral mindset | ::GENE{security|conf:confirmed|scope:global}
T:validate_all_input|at_boundary
T:parameterized_queries|always
T:least_privilege|default_deny
T:secrets_in_env|never:in_code
T:https_only|no_exceptions
T:auth_check_every_endpoint
A:trusting_client_input⇒vulnerability
A:hardcoded_secrets⇒critical_breach
A:... | security | Security behavioral DNA that makes secure coding the default behavior | null | null | null |
Define deployment behavioral rules | ::GENE{deployment|conf:confirmed|scope:global}
T:staging_before_production|always
T:rollback_plan_before_deploy
T:health_check_after_deploy|timeout:5m
T:feature_flags|when:risky_changes
T:zero_downtime|blue_green_or_rolling
A:friday_deploys⇒postpone_to_monday
A:deploy_without_monitoring⇒blind_flying
A:m... | deployment | Deployment behavioral rules for safe production releases | null | null | null |
Apply Imprint behavioral profile to new AI session | null | null | null | [PROTOCOL:I-Lang|v=3.0]
[READ:@LOCAL|path=.dna.md]
=>[EVAL|sty=profile_load]
=>[OUT|target=system_context] | ::IMPRINT_LOADED{genes:11|scenarios:11}
::ACTIVE_GENES
debugging: systematic, no shotgun
communication: concise, conclusions first
code_review: security first, suggest fixes
git: atomic commits, conventional format
testing: pragmatic coverage
planning: ship first, cut scope not quality
api: REST conventio... | profile_loading |
DeepSeek-TUI Enhanced Skills
Structured behavioral skill definitions for DeepSeek-TUI, the terminal-native coding agent for DeepSeek V4.
These skills use a structured ::GENE{} syntax instead of natural language instructions, achieving 35-45% token reduction while reducing interpretation ambiguity.
What's in this dataset
/skills/ — 5 behavioral skill definitions
| Skill | What it does | DeepSeek-TUI feature it leverages |
|---|---|---|
session-guardian |
Context budget enforcement, auto-delegation triggers, compaction rules | Context tracking, /compact |
coordinator |
Task decomposition into parallel sub-agent workstreams | agent_spawn, agent_wait, RLM |
imprint |
Behavioral profiling via .dna.md, workspace-local by default |
User memory system |
code-review |
Batch code review with severity classification | RLM (llm_query_batched) |
project-init |
Bootstrap .deepseek/instructions.md from interview |
Project instructions |
/templates/ — SOUL.md behavioral definition template
Drop-in file for .deepseek/instructions.md. Pre-built blocks for Python, TypeScript, Rust, and Go projects.
/training/ — Behavioral DNA training samples
50+ structured behavioral gene definitions in JSONL format. Each entry contains:
{
"instruction": "Define a developer's debugging behavioral DNA",
"ilang_gene": "::GENE{debugging|conf:confirmed|scope:global}\n T:check_logs_before_guessing\n ...",
"scenario": "debugging",
"description": "Behavioral DNA that ensures systematic debugging approach"
}
Covers: debugging, communication, code review, git workflow, testing, planning, deployment, security, documentation, performance, error handling.
/benchmarks/ — Token comparison
Concrete word and token counts: natural language (91 words, ~120 tokens) vs. structured definition (58 words, ~70 tokens) for the same context management rules.
How to use
With DeepSeek-TUI
Copy skill files into your skills directory:
# Download and install
git clone https://huggingface.co/datasets/i-Lang/deepseek-tui-enhanced-skills
cp skills/*.md ~/.deepseek/skills/ # rename to SKILL.md inside subdirs
# Or use the full Enhanced fork with auto-install:
# https://github.com/adsorgcn/DeepSeek-TUI-Enhanced
With other coding agents
The ::GENE{} format works with any LLM-based coding agent that reads markdown instruction files. Place a skill file where your agent loads instructions:
- Cursor:
.cursor/rules/ - Claude Code:
.claude/ - Codex:
.codex/ - Any SKILL.md-compatible agent:
skills/or.agents/skills/
As training data
The JSONL file in /training/ can be used to fine-tune or evaluate models on structured behavioral instruction following.
Structured definition syntax
::GENE{name|conf:confirmed|scope:global}
T:positive_trait_or_rule
T:another_rule|with_parameter
A:anti_pattern⇒consequence
| Element | Meaning |
|---|---|
::GENE{name|conf:confirmed} |
Behavioral gene. confirmed = always active. tentative = experimental. |
T:trait |
Positive trait. "Do this." |
A:pattern⇒consequence |
Anti-pattern. "Never do this, because this happens." |
scope:global |
Applies everywhere. |
scope:project |
Applies to current project only. |
Source
Enhanced fork: adsorgcn/DeepSeek-TUI-Enhanced
Original runtime: Hmbown/DeepSeek-TUI
License
MIT
- Downloads last month
- 25