codehava-agent-kit 1.0.1 → 3.0.0

package/templates/.agent/skills/context-agent/scripts/session_summary.py

@@ -0,0 +1,319 @@
+"""
+Gerador de resumos estruturados de sessão.
+Analisa mensagens e gera session-NNN.md.
+"""
+
+import re
+from datetime import datetime
+from pathlib import Path
+
+from config import (
+    SESSIONS_DIR,
+    DECISION_MARKERS,
+    PENDING_MARKERS,
+)
+from models import SessionSummary, PendingTask, SessionEntry
+
+
+def get_next_session_number() -> int:
+    """Retorna o próximo número de sessão disponível."""
+    SESSIONS_DIR.mkdir(parents=True, exist_ok=True)
+    existing = list(SESSIONS_DIR.glob("session-*.md"))
+    if not existing:
+        return 1
+    numbers = []
+    for f in existing:
+        try:
+            num = int(f.stem.split("-")[1])
+            numbers.append(num)
+        except (IndexError, ValueError):
+            continue
+    return max(numbers) + 1 if numbers else 1
+
+
+def generate_summary(
+    entries: list[SessionEntry],
+    session_number: int,
+    metadata: dict,
+) -> SessionSummary:
+    """Gera um resumo estruturado a partir das entradas da sessão."""
+    user_messages = [e.content for e in entries if e.role == "user" and e.content.strip()]
+    assistant_messages = [e.content for e in entries if e.role == "assistant" and e.content.strip()]
+    all_messages = user_messages + assistant_messages
+    all_tool_calls = []
+    all_files_modified = []
+
+    for e in entries:
+        all_tool_calls.extend(e.tool_calls)
+        all_files_modified.extend(e.files_modified)
+
+    # Deduplicate files
+    seen_files = set()
+    unique_files = []
+    for f in all_files_modified:
+        if f["path"] not in seen_files:
+            seen_files.add(f["path"])
+            unique_files.append(f)
+
+    # Extrair data
+    date_str = ""
+    start_time = metadata.get("start_time", "")
+    if start_time:
+        try:
+            dt = datetime.fromisoformat(start_time.replace("Z", "+00:00"))
+            date_str = dt.strftime("%Y-%m-%d")
+        except ValueError:
+            date_str = datetime.now().strftime("%Y-%m-%d")
+    else:
+        date_str = datetime.now().strftime("%Y-%m-%d")
+
+    summary = SessionSummary(
+        session_number=session_number,
+        session_id=metadata.get("session_id", ""),
+        slug=metadata.get("slug", ""),
+        date=date_str,
+        start_time=start_time,
+        end_time=metadata.get("end_time", ""),
+        duration_minutes=metadata.get("duration_minutes", 0),
+        model=metadata.get("model", ""),
+        total_input_tokens=metadata.get("total_input_tokens", 0),
+        total_output_tokens=metadata.get("total_output_tokens", 0),
+        total_cache_tokens=metadata.get("total_cache_tokens", 0),
+        message_count=metadata.get("message_count", 0),
+        tool_call_count=metadata.get("tool_call_count", 0),
+        files_modified=unique_files,
+    )
+
+    # Extrair tópicos das mensagens do usuário
+    summary.topics = _extract_topics(user_messages)
+
+    # Extrair decisões
+    summary.decisions = _extract_decisions(all_messages)
+
+    # Extrair tarefas
+    summary.tasks_completed = _extract_completed_tasks(all_messages)
+    summary.tasks_pending = _extract_pending_tasks(all_messages, session_number, date_str)
+
+    # Extrair erros
+    summary.errors_resolved = _extract_errors(assistant_messages)
+
+    # Extrair findings
+    summary.key_findings = _extract_findings(assistant_messages)
+
+    return summary
+
+
+def _extract_topics(user_messages: list[str]) -> list[str]:
+    """Identifica tópicos principais das mensagens do usuário."""
+    topics = []
+    for msg in user_messages:
+        # Limpar mensagens muito longas
+        msg_clean = msg[:500] if len(msg) > 500 else msg
+        # Pegar a primeira frase significativa como tópico
+        sentences = re.split(r'[.!?\n]', msg_clean)
+        for s in sentences:
+            s = s.strip()
+            if len(s) > 10 and len(s) < 200:
+                topics.append(s)
+                break
+
+    # Deduplicate e limitar
+    seen = set()
+    unique = []
+    for t in topics:
+        t_lower = t.lower()
+        if t_lower not in seen:
+            seen.add(t_lower)
+            unique.append(t)
+
+    return unique[:10]
+
+
+def _extract_decisions(messages: list[str]) -> list[str]:
+    """Encontra decisões nas mensagens."""
+    decisions = []
+    for msg in messages:
+        for line in msg.split("\n"):
+            line_lower = line.lower().strip()
+            for marker in DECISION_MARKERS:
+                if marker in line_lower:
+                    clean = line.strip()
+                    if 15 < len(clean) < 300:
+                        decisions.append(clean)
+                    break
+    return list(dict.fromkeys(decisions))[:10]  # Deduplicate, max 10
+
+
+def _extract_completed_tasks(messages: list[str]) -> list[str]:
+    """Encontra tarefas concluídas."""
+    completed = []
+    patterns = [
+        r"- \[x\]\s+(.+)",
+        r"✅\s+(.+)",
+        r"(?:concluí|completei|terminei|finalizei|done|completed|finished)\s+(.+)",
+    ]
+    for msg in messages:
+        for line in msg.split("\n"):
+            for pattern in patterns:
+                m = re.search(pattern, line, re.IGNORECASE)
+                if m:
+                    task = m.group(1).strip()
+                    if len(task) > 5:
+                        completed.append(task)
+    return list(dict.fromkeys(completed))[:15]
+
+
+def _extract_pending_tasks(
+    messages: list[str],
+    session_number: int,
+    date: str,
+) -> list[PendingTask]:
+    """Encontra tarefas pendentes. Foca em checkboxes não marcados nas mensagens do user."""
+    tasks = []
+    for msg in messages:
+        # Ignorar mensagens muito longas (provavelmente tool results, não conversação)
+        if len(msg) > 5000:
+            continue
+
+        for line in msg.split("\n"):
+            stripped = line.strip()
+
+            # Checkbox não marcado — sinal claro de tarefa
+            m = re.match(r"- \[ \]\s+(.+)", stripped)
+            if m:
+                desc = m.group(1).strip()
+                # Filtrar descrições que parecem código/documentação
+                if 10 < len(desc) < 200 and not desc.startswith("`") and not desc.startswith("-"):
+                    tasks.append(PendingTask(
+                        description=desc,
+                        source_session=session_number,
+                        created_date=date,
+                    ))
+
+    # Deduplicate
+    seen = set()
+    unique = []
+    for t in tasks:
+        if t.description not in seen:
+            seen.add(t.description)
+            unique.append(t)
+
+    return unique[:10]
+
+
+def _extract_errors(assistant_messages: list[str]) -> list[dict]:
+    """Encontra erros e suas soluções."""
+    errors = []
+    error_patterns = [
+        r"(?:error|erro|falha|failed|exception)[\s:]+(.+)",
+    ]
+    for msg in assistant_messages:
+        for pattern in error_patterns:
+            matches = re.findall(pattern, msg, re.IGNORECASE)
+            for match in matches:
+                if len(match) > 10:
+                    errors.append({"error": match[:200], "solution": ""})
+    return errors[:5]
+
+
+def _extract_findings(assistant_messages: list[str]) -> list[str]:
+    """Extrai descobertas/findings importantes."""
+    findings = []
+    markers = [
+        "descobri que", "encontrei", "notei que", "importante:",
+        "found that", "noticed that", "important:", "key finding",
+    ]
+    for msg in assistant_messages:
+        for line in msg.split("\n"):
+            line_lower = line.lower().strip()
+            for marker in markers:
+                if marker in line_lower and len(line.strip()) > 20:
+                    findings.append(line.strip()[:200])
+                    break
+    return list(dict.fromkeys(findings))[:5]
+
+
+def save_session_summary(summary: SessionSummary):
+    """Salva resumo como arquivo markdown."""
+    SESSIONS_DIR.mkdir(parents=True, exist_ok=True)
+    path = SESSIONS_DIR / f"session-{summary.session_number:03d}.md"
+
+    lines = [
+        f"# Sessão {summary.session_number:03d} — {summary.date}",
+        f"**Slug:** {summary.slug} | **Duração:** ~{summary.duration_minutes}min | **Modelo:** {summary.model}",
+        "",
+    ]
+
+    if summary.topics:
+        lines.append("## Tópicos")
+        for t in summary.topics:
+            lines.append(f"- {t}")
+        lines.append("")
+
+    if summary.decisions:
+        lines.append("## Decisões")
+        for d in summary.decisions:
+            lines.append(f"- {d}")
+        lines.append("")
+
+    if summary.tasks_completed:
+        lines.append("## Tarefas Concluídas")
+        for t in summary.tasks_completed:
+            lines.append(f"- [x] {t}")
+        lines.append("")
+
+    if summary.tasks_pending:
+        lines.append("## Tarefas Pendentes")
+        for t in summary.tasks_pending:
+            if isinstance(t, PendingTask):
+                lines.append(f"- [ ] {t.description} (prioridade: {t.priority})")
+            else:
+                lines.append(f"- [ ] {t}")
+        lines.append("")
+
+    if summary.files_modified:
+        lines.append("## Arquivos Modificados")
+        for f in summary.files_modified:
+            lines.append(f"- `{f['path']}` — {f['action']}")
+        lines.append("")
+
+    if summary.key_findings:
+        lines.append("## Descobertas")
+        for f in summary.key_findings:
+            lines.append(f"- {f}")
+        lines.append("")
+
+    if summary.errors_resolved:
+        lines.append("## Erros Resolvidos")
+        for e in summary.errors_resolved:
+            lines.append(f"- {e['error']}")
+        lines.append("")
+
+    if summary.open_questions:
+        lines.append("## Questões em Aberto")
+        for q in summary.open_questions:
+            lines.append(f"- {q}")
+        lines.append("")
+
+    if summary.technical_debt:
+        lines.append("## Dívida Técnica")
+        for d in summary.technical_debt:
+            lines.append(f"- {d}")
+        lines.append("")
+
+    lines.append("## Métricas")
+    lines.append(f"- Input tokens: {summary.total_input_tokens:,}")
+    lines.append(f"- Output tokens: {summary.total_output_tokens:,}")
+    lines.append(f"- Cache tokens: {summary.total_cache_tokens:,}")
+    lines.append(f"- Mensagens: {summary.message_count}")
+    lines.append(f"- Tool calls: {summary.tool_call_count}")
+    lines.append("")
+
+    # Link para sessão anterior
+    if summary.session_number > 1:
+        prev = summary.session_number - 1
+        lines.append("---")
+        lines.append(f"*Sessão anterior: [session-{prev:03d}](session-{prev:03d}.md)*")
+
+    path.write_text("\n".join(lines), encoding="utf-8")
+    return path

package/templates/.agent/skills/context-optimization/SKILL.md

@@ -0,0 +1,180 @@
+---
+name: context-optimization
+description: "Context optimization extends the effective capacity of limited context windows through strategic compression, masking, caching, and partitioning. The goal is not to magically increase context windows but to make better use of available capacity."
+risk: unknown
+source: community
+---
+
+# Context Optimization Techniques
+
+Context optimization extends the effective capacity of limited context windows through strategic compression, masking, caching, and partitioning. The goal is not to magically increase context windows but to make better use of available capacity. Effective optimization can double or triple effective context capacity without requiring larger models or longer contexts.
+
+## When to Use
+Activate this skill when:
+- Context limits constrain task complexity
+- Optimizing for cost reduction (fewer tokens = lower costs)
+- Reducing latency for long conversations
+- Implementing long-running agent systems
+- Needing to handle larger documents or conversations
+- Building production systems at scale
+
+## Core Concepts
+
+Context optimization extends effective capacity through four primary strategies: compaction (summarizing context near limits), observation masking (replacing verbose outputs with references), KV-cache optimization (reusing cached computations), and context partitioning (splitting work across isolated contexts).
+
+The key insight is that context quality matters more than quantity. Optimization preserves signal while reducing noise. The art lies in selecting what to keep versus what to discard, and when to apply each technique.
+
+## Detailed Topics
+
+### Compaction Strategies
+
+**What is Compaction**
+Compaction is the practice of summarizing context contents when approaching limits, then reinitializing a new context window with the summary. This distills the contents of a context window in a high-fidelity manner, enabling the agent to continue with minimal performance degradation.
+
+Compaction typically serves as the first lever in context optimization. The art lies in selecting what to keep versus what to discard.
+
+**Compaction Implementation**
+Compaction works by identifying sections that can be compressed, generating summaries that capture essential points, and replacing full content with summaries. Priority for compression goes to tool outputs (replace with summaries), old turns (summarize early conversation), retrieved docs (summarize if recent versions exist), and never compress system prompt.
+
+**Summary Generation**
+Effective summaries preserve different elements depending on message type:
+
+Tool outputs: Preserve key findings, metrics, and conclusions. Remove verbose raw output.
+
+Conversational turns: Preserve key decisions, commitments, and context shifts. Remove filler and back-and-forth.
+
+Retrieved documents: Preserve key facts and claims. Remove supporting evidence and elaboration.
+
+### Observation Masking
+
+**The Observation Problem**
+Tool outputs can comprise 80%+ of token usage in agent trajectories. Much of this is verbose output that has already served its purpose. Once an agent has used a tool output to make a decision, keeping the full output provides diminishing value while consuming significant context.
+
+Observation masking replaces verbose tool outputs with compact references. The information remains accessible if needed but does not consume context continuously.
+
+**Masking Strategy Selection**
+Not all observations should be masked equally:
+
+Never mask: Observations critical to current task, observations from the most recent turn, observations used in active reasoning.
+
+Consider masking: Observations from 3+ turns ago, verbose outputs with key points extractable, observations whose purpose has been served.
+
+Always mask: Repeated outputs, boilerplate headers/footers, outputs already summarized in conversation.
+
+### KV-Cache Optimization
+
+**Understanding KV-Cache**
+The KV-cache stores Key and Value tensors computed during inference, growing linearly with sequence length. Caching the KV-cache across requests sharing identical prefixes avoids recomputation.
+
+Prefix caching reuses KV blocks across requests with identical prefixes using hash-based block matching. This dramatically reduces cost and latency for requests with common prefixes like system prompts.
+
+**Cache Optimization Patterns**
+Optimize for caching by reordering context elements to maximize cache hits. Place stable elements first (system prompt, tool definitions), then frequently reused elements, then unique elements last.
+
+Design prompts to maximize cache stability: avoid dynamic content like timestamps, use consistent formatting, keep structure stable across sessions.
+
+### Context Partitioning
+
+**Sub-Agent Partitioning**
+The most aggressive form of context optimization is partitioning work across sub-agents with isolated contexts. Each sub-agent operates in a clean context focused on its subtask without carrying accumulated context from other subtasks.
+
+This approach achieves separation of concerns—the detailed search context remains isolated within sub-agents while the coordinator focuses on synthesis and analysis.
+
+**Result Aggregation**
+Aggregate results from partitioned subtasks by validating all partitions completed, merging compatible results, and summarizing if still too large.
+
+### Budget Management
+
+**Context Budget Allocation**
+Design explicit context budgets. Allocate tokens to categories: system prompt, tool definitions, retrieved docs, message history, and reserved buffer. Monitor usage against budget and trigger optimization when approaching limits.
+
+**Trigger-Based Optimization**
+Monitor signals for optimization triggers: token utilization above 80%, degradation indicators, and performance drops. Apply appropriate optimization techniques based on context composition.
+
+## Practical Guidance
+
+### Optimization Decision Framework
+
+When to optimize:
+- Context utilization exceeds 70%
+- Response quality degrades as conversations extend
+- Costs increase due to long contexts
+- Latency increases with conversation length
+
+What to apply:
+- Tool outputs dominate: observation masking
+- Retrieved documents dominate: summarization or partitioning
+- Message history dominates: compaction with summarization
+- Multiple components: combine strategies
+
+### Performance Considerations
+
+Compaction should achieve 50-70% token reduction with less than 5% quality degradation. Masking should achieve 60-80% reduction in masked observations. Cache optimization should achieve 70%+ hit rate for stable workloads.
+
+Monitor and iterate on optimization strategies based on measured effectiveness.
+
+## Examples
+
+**Example 1: Compaction Trigger**
+```python
+if context_tokens / context_limit > 0.8:
+    context = compact_context(context)
+```
+
+**Example 2: Observation Masking**
+```python
+if len(observation) > max_length:
+    ref_id = store_observation(observation)
+    return f"[Obs:{ref_id} elided. Key: {extract_key(observation)}]"
+```
+
+**Example 3: Cache-Friendly Ordering**
+```python
+# Stable content first
+context = [system_prompt, tool_definitions]  # Cacheable
+context += [reused_templates]  # Reusable
+context += [unique_content]  # Unique
+```
+
+## Guidelines
+
+1. Measure before optimizing—know your current state
+2. Apply compaction before masking when possible
+3. Design for cache stability with consistent prompts
+4. Partition before context becomes problematic
+5. Monitor optimization effectiveness over time
+6. Balance token savings against quality preservation
+7. Test optimization at production scale
+8. Implement graceful degradation for edge cases
+
+## Integration
+
+This skill builds on context-fundamentals and context-degradation. It connects to:
+
+- multi-agent-patterns - Partitioning as isolation
+- evaluation - Measuring optimization effectiveness
+- memory-systems - Offloading context to memory
+
+## References
+
+Internal reference:
+- Optimization Techniques Reference - Detailed technical reference
+
+Related skills in this collection:
+- context-fundamentals - Context basics
+- context-degradation - Understanding when to optimize
+- evaluation - Measuring optimization
+
+External resources:
+- Research on context window limitations
+- KV-cache optimization techniques
+- Production engineering guides
+
+---
+
+## Skill Metadata
+
+**Created**: 2025-12-20
+**Last Updated**: 2025-12-20
+**Author**: Agent Skills for Context Engineering Contributors
+**Version**: 1.0.0