Commit Graph

11 Commits

Author SHA1 Message Date
YeonGyu-Kim e87579b687 feat(agents): wire Gemini prompt routing into Sisyphus-Junior, Atlas, Prometheus
Add 'gemini' to prompt source types and route Gemini models to new
Gemini-optimized prompts via isGeminiModel detection. Update barrel
exports for all 3 agent modules. All existing tests pass.
2026-02-22 15:30:40 +09:00
YeonGyu-Kim 0425c75447 feat(agents): add Gemini-optimized prompts for Sisyphus, Sisyphus-Junior, Prometheus, Atlas
Gemini models are aggressively optimistic and avoid tool calls in favor of
internal reasoning. These prompts counter that with:
- TOOL_CALL_MANDATE sections forcing actual tool usage
- Anti-optimism checkpoints before claiming completion
- Stronger delegation enforcement (Gemini prefers doing work itself)
- Aggressive verification language (subagent results are 'EXTREMELY SUSPICIOUS')
- Mandatory thinking checkpoints in Prometheus (prevents jumping to conclusions)
- Scope discipline reminders (creativity → implementation quality, not scope creep)
2026-02-22 15:30:40 +09:00
YeonGyu-Kim 80bb262066 refactor(agents): convert all markdown tables to bullet lists across 12 agent files
Tables in template literal prompts render poorly in some LLM contexts.
Replaced 43 table instances with equivalent bullet list format preserving
all information. Affected: hephaestus, atlas/default, atlas/prompt-section-builder,
sisyphus-junior/gpt, librarian, explore, metis, prometheus/behavioral-summary,
prometheus/identity-constraints, prometheus/interview-mode, prometheus/plan-generation,
prometheus/plan-template.
2026-02-18 21:10:21 +09:00
YeonGyu-Kim 8ea24d26d2 chore: upgrade claude-sonnet-4-5 to claude-sonnet-4-6 across codebase 2026-02-18 15:51:24 +09:00
YeonGyu-Kim 2287392203 update: Hephaestus completion guarantee, Sisyphus-Junior Hephaestus-style rewrite, snake_case tools
Hephaestus:
- Add Completion Guarantee section with Codex-style persistence framing
- Add explicit explore/librarian call syntax examples (subagent_type, not category)
- Use positive 'keep going until resolved' over negative 'NEVER stop'
- Fix tool names: TaskCreate/TaskUpdate → task_create/task_update

Sisyphus-Junior GPT:
- Full Hephaestus-style rewrite: autonomy, reporting, parallelism, tool usage
- Remove Blocked & Allowed Tools section and 'You work ALONE' messaging
- Add Progress Updates, Ambiguity Protocol, Code Quality sections
- Fix tool names: TaskCreate/TaskUpdate → task_create/task_update

Sisyphus-Junior Default:
- Remove buildConstraintsSection and blocked actions messaging
- Fix tool names: TaskCreate/TaskUpdate → task_create/task_update

Tests: update all assertions for new prompt structure (31/31 pass)
2026-02-17 03:12:32 +09:00
YeonGyu-Kim e9bc78b9a0 feat(agents): add file:// URI support in prompt_append configuration
Port devxoul's PR #821 feature to current codebase structure.
Supports absolute, relative, ~/home paths with percent-encoding.
Gracefully handles malformed URIs and missing files with warnings.

Co-authored-by: devxoul <devxoul@gmail.com>
2026-02-13 11:25:40 +09:00
YeonGyu-Kim bce3732222 fix(sisyphus-junior): disambiguate blocked delegation tool from allowed task management tools
When task_system is enabled, the prompt said 'task tool: BLOCKED' which
LLMs interpreted as blocking task_create/task_update/task_list/task_get
too. Now the constraints section explicitly separates 'task (agent
delegation tool): BLOCKED' from 'task_create, task_update, ...: ALLOWED'
so Junior no longer refuses to use task management tools.
2026-02-10 11:41:44 +09:00
YeonGyu-Kim 5db49e722f fix(agent-config): pass useTaskSystem to sisyphus-junior when task_system is enabled
sisyphus-junior prompt always used todo-based discipline text regardless of
experimental.task_system setting because the useTaskSystem flag was never
forwarded from agent-config-handler to createSisyphusJuniorAgentWithOverrides.
2026-02-09 21:10:15 +09:00
YeonGyu-Kim df03300211 refactor: wave 1 - extract leaf modules, rename catch-all files, split index.ts hooks
- Split 25+ index.ts files into hook.ts + extracted modules
- Rename all catch-all utils.ts/helpers.ts to domain-specific names
- Split src/tools/lsp/ into ~15 focused modules
- Split src/tools/delegate-task/ into ~18 focused modules
- Separate shared types from implementation
- 155 files changed, 60+ new files created
- All typecheck clean, 61 tests pass
2026-02-08 13:57:26 +09:00
YeonGyu-Kim b30a131690 refactor: migrate delegate_task to task tool with metadata fixes
- Rename delegate_task tool to task across codebase (100 files)
- Update model references: claude-opus-4-6 → 4-5, gpt-5.3-codex → 5.2-codex
- Add tool-metadata-store to restore metadata overwritten by fromPlugin()
- Add session ID polling for BackgroundManager task sessions
- Await async ctx.metadata() calls in tool executors
- Add ses_ prefix guard to getMessageDir for performance
- Harden BackgroundManager with idle deferral and error handling
- Fix duplicate task key in sisyphus-junior test object literals
- Fix unawaited showOutputToUser in ast_grep_replace
- Fix background=true → run_in_background=true in ultrawork prompt
- Fix duplicate task/task references in docs and comments
2026-02-06 21:35:30 +09:00
YeonGyu-Kim f787e15dc9 feat(agents): add GPT-5.2 optimized prompt for sisyphus-junior
Restructure sisyphus-junior agent to use model-specific prompts:
- GPT models: GPT-5.2 prompting guide principles (verbosity constraints,
  scope discipline, tool usage rules, explicit decision criteria)
- Claude models: Original prompt with extended reasoning context

Directory structure now mirrors atlas/ pattern for consistency.
2026-02-03 14:12:22 +09:00