@pennixrv/trellis 0.6.31 → 0.6.34

package/dist/templates/trellis/scripts/common/context_projection.py

@@ -0,0 +1,508 @@
+#!/usr/bin/env python3
+"""Shared materialization for Python sub-agent context.
+
+The shared Python Hook renders this projection for implement/check agents and
+``task.py validate`` reads the same result before a task starts.  A notice is
+still useful in a live prompt, but it is never a complete projection.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from dataclasses import dataclass, field
+from pathlib import Path
+
+from .paths import DIR_ARCHIVE, DIR_TASKS, DIR_WORKFLOW
+
+
+DIRECTORY_MAX_FILES = 20
+
+
+@dataclass(frozen=True)
+class ContextEntry:
+    """A normalised JSONL entry consumed by the Python Hook."""
+
+    path: str
+    entry_type: str
+    reason: str
+    line: int
+
+
+@dataclass(frozen=True)
+class ManifestProblem:
+    """A malformed manifest row that the CLI must reject."""
+
+    line: int
+    message: str
+
+
+@dataclass(frozen=True)
+class ProjectionIssue:
+    """Why one declared source was not represented as complete body text."""
+
+    role: str
+    category: str
+    path: str
+    reason: str
+    line: int | None = None
+
+
+@dataclass
+class ContextProjection:
+    """The Hook text and the facts the validator must fail closed on."""
+
+    role: str
+    text: str
+    entries: list[ContextEntry] = field(default_factory=list)
+    manifest_problems: list[ManifestProblem] = field(default_factory=list)
+    issues: list[ProjectionIssue] = field(default_factory=list)
+    manifest_exists: bool = False
+
+
+class _Budget:
+    def __init__(self, max_total_bytes: int) -> None:
+        self.max_total_bytes = max_total_bytes
+        self.used = 0
+
+    def has_room(self, size: int) -> bool:
+        return self.max_total_bytes <= 0 or self.used + size <= self.max_total_bytes
+
+    def add(self, size: int) -> None:
+        self.used += size
+
+
+def truncate_utf8(data: bytes, cap: int) -> bytes:
+    """Truncate at a valid UTF-8 boundary; zero means unlimited."""
+    if cap <= 0 or len(data) <= cap:
+        return data
+
+    truncated = data[:cap]
+    index = len(truncated)
+    while index > 0 and (truncated[index - 1] & 0xC0) == 0x80:
+        index -= 1
+    if index == 0:
+        return b""
+
+    lead = truncated[index - 1]
+    if lead & 0x80:
+        if (lead & 0xE0) == 0xC0:
+            sequence_length = 2
+        elif (lead & 0xF0) == 0xE0:
+            sequence_length = 3
+        elif (lead & 0xF8) == 0xF0:
+            sequence_length = 4
+        else:
+            sequence_length = 1
+        if (index - 1) + sequence_length > len(truncated):
+            index -= 1
+    return truncated[:index]
+
+
+def is_binary_content(data: bytes) -> bool:
+    if b"\x00" in data:
+        return True
+    try:
+        data.decode("utf-8", errors="strict")
+    except UnicodeDecodeError:
+        return True
+    return False
+
+
+def parse_context_manifest(jsonl_file: Path) -> tuple[list[ContextEntry], list[ManifestProblem], bool]:
+    """Parse the JSONL schema shared by the Hook and validation command."""
+    if not jsonl_file.is_file():
+        return [], [], False
+    try:
+        lines = jsonl_file.read_text(encoding="utf-8").splitlines()
+    except (OSError, UnicodeDecodeError):
+        return [], [ManifestProblem(0, "Manifest is unreadable or is not valid UTF-8")], True
+
+    entries: list[ContextEntry] = []
+    problems: list[ManifestProblem] = []
+    for line_number, line in enumerate(lines, start=1):
+        if not line.strip():
+            continue
+        try:
+            data = json.loads(line)
+        except json.JSONDecodeError:
+            problems.append(ManifestProblem(line_number, "Invalid JSON"))
+            continue
+        if not isinstance(data, dict):
+            problems.append(ManifestProblem(line_number, "Expected a JSON object"))
+            continue
+        if "_example" in data:
+            problems.append(
+                ManifestProblem(
+                    line_number,
+                    "Placeholder `_example` row left by an older task.py create — delete this line, or replace it with "
+                    '{"file": "<path>", "reason": "<why>"}',
+                )
+            )
+            continue
+
+        file_value = data.get("file")
+        path_value = data.get("path")
+        if file_value not in (None, "") and not isinstance(file_value, str):
+            problems.append(
+                ManifestProblem(line_number, "`file` or legacy `path` must be a string path")
+            )
+            continue
+        if path_value not in (None, "") and not isinstance(path_value, str):
+            problems.append(
+                ManifestProblem(line_number, "`file` or legacy `path` must be a string path")
+            )
+            continue
+        value = file_value or path_value
+        if not value:
+            continue
+        if not isinstance(value, str):
+            problems.append(ManifestProblem(line_number, "`file` or legacy `path` must be a string path"))
+            continue
+
+        entry_type = data.get("type", "file")
+        reason = data.get("reason") or "-"
+        entries.append(
+            ContextEntry(
+                path=value,
+                entry_type=entry_type if isinstance(entry_type, str) else "file",
+                reason=reason if isinstance(reason, str) else str(reason),
+                line=line_number,
+            )
+        )
+    return entries, problems, True
+
+
+def _real_path_contained(base_real: str, target_real: str) -> bool:
+    try:
+        return os.path.commonpath([base_real, target_real]) == base_real
+    except ValueError:
+        return False
+
+
+def _is_allowed_path(repo_root: Path, candidate: Path) -> bool:
+    try:
+        root_real = os.path.realpath(repo_root)
+        workflow_real = os.path.realpath(repo_root / DIR_WORKFLOW)
+        candidate_real = os.path.realpath(candidate)
+    except OSError:
+        return False
+    return _real_path_contained(root_real, candidate_real) or _real_path_contained(
+        workflow_real, candidate_real
+    )
+
+
+def resolve_context_entry_path(file_path: str, repo_root: Path, task_dir: Path | None) -> Path | None:
+    """Resolve an entry and preserve archived task self-reference semantics."""
+    repo_path = repo_root / file_path
+    if task_dir is None:
+        return repo_path
+
+    try:
+        task_parts = task_dir.resolve().relative_to(repo_root.resolve()).parts
+    except ValueError:
+        return repo_path
+
+    archive_prefix = (DIR_WORKFLOW, DIR_TASKS, DIR_ARCHIVE)
+    if len(task_parts) != 5 or task_parts[:3] != archive_prefix:
+        return repo_path
+
+    year_month = task_parts[3]
+    if (
+        len(year_month) != 7
+        or year_month[4] != "-"
+        or not year_month[:4].isdigit()
+        or not year_month[5:].isdigit()
+    ):
+        return repo_path
+
+    historical_root = f"{DIR_WORKFLOW}/{DIR_TASKS}/{task_dir.name}"
+    posix_path = file_path.replace("\\", "/")
+    if posix_path == historical_root:
+        relative_parts: tuple[str, ...] = ()
+    elif posix_path.startswith(f"{historical_root}/"):
+        relative_path = posix_path[len(historical_root) + 1 :].rstrip("/")
+        relative_parts = tuple(relative_path.split("/")) if relative_path else ()
+        if any(part in ("", ".", "..") for part in relative_parts):
+            return None
+    else:
+        return repo_path
+
+    try:
+        archive_root = task_dir.resolve()
+        resolved_path = task_dir.joinpath(*relative_parts).resolve()
+        resolved_path.relative_to(archive_root)
+    except (OSError, RuntimeError, ValueError):
+        return None
+    return resolved_path
+
+
+def _display_task_path(repo_root: Path, task_dir: Path, file_name: str) -> str:
+    try:
+        return (task_dir.relative_to(repo_root) / file_name).as_posix()
+    except ValueError:
+        return f"{task_dir.as_posix().rstrip('/')}/{file_name}"
+
+
+def _truncate_notice(path: str, cap: int) -> str:
+    return f"\n[Trellis: truncated at {cap} bytes — read {path} for the full content]"
+
+
+def _binary_notice(path: str, size: int, reason: str) -> str:
+    return f"[Trellis: not inlined (binary file) — {path} ({size} bytes): {reason}]"
+
+
+def _index_notice(path: str, size: int, reason: str) -> str:
+    return f"[Trellis: not inlined (total context limit reached) — {path} ({size} bytes): {reason}]"
+
+
+def _append_block(
+    projection: ContextProjection,
+    budget: _Budget,
+    *,
+    header: str,
+    path: str,
+    content: str,
+    size: int,
+    reason: str,
+    category: str,
+    line: int | None,
+) -> str:
+    block = f"=== {header} ===\n{content}"
+    block_bytes = len(block.encode("utf-8"))
+    if budget.has_room(block_bytes):
+        budget.add(block_bytes)
+        return block
+
+    notice = _index_notice(path, size, reason)
+    budget.add(len(notice.encode("utf-8")))
+    limit = budget.max_total_bytes
+    projection.issues.append(
+        ProjectionIssue(
+            role=projection.role,
+            category="total",
+            path=path,
+            reason=f"exceeds context_injection.max_total_bytes ({limit}); injection uses an index notice",
+            line=line,
+        )
+    )
+    return notice
+
+
+def _new_projection(role: str, entries: list[ContextEntry], problems: list[ManifestProblem], manifest_exists: bool) -> ContextProjection:
+    projection = ContextProjection(
+        role=role,
+        text="",
+        entries=entries,
+        manifest_problems=problems,
+        manifest_exists=manifest_exists,
+    )
+    return projection
+
+
+def _read_material(
+    repo_root: Path, target: Path | None, category: str, path: str, projection: ContextProjection, line: int | None
+) -> bytes | None:
+    if target is None:
+        projection.issues.append(
+            ProjectionIssue(projection.role, category, path, "path cannot be resolved safely", line)
+        )
+        return None
+    if not _is_allowed_path(repo_root, target):
+        projection.issues.append(
+            ProjectionIssue(projection.role, category, path, "path escapes the allowed project roots", line)
+        )
+        return None
+    try:
+        if not target.is_file():
+            raise OSError
+        return target.read_bytes()
+    except OSError:
+        projection.issues.append(
+            ProjectionIssue(projection.role, category, path, "file is missing or unreadable", line)
+        )
+        return None
+
+
+def _materialize_file(
+    repo_root: Path,
+    target: Path | None,
+    path: str,
+    reason: str,
+    limits: dict[str, int],
+    budget: _Budget,
+    projection: ContextProjection,
+    *,
+    category: str,
+    line: int | None,
+    header: str | None = None,
+    cap_key: str = "max_file_bytes",
+) -> str | None:
+    data = _read_material(repo_root, target, category, path, projection, line)
+    if data is None:
+        return None
+
+    size = len(data)
+    if is_binary_content(data):
+        notice = _binary_notice(path, size, reason)
+        budget.add(len(notice.encode("utf-8")))
+        projection.issues.append(
+            ProjectionIssue(projection.role, category, path, "binary content is not inlined", line)
+        )
+        return notice
+
+    cap = limits[cap_key]
+    materialized = truncate_utf8(data, cap)
+    content = materialized.decode("utf-8")
+    if len(materialized) < size:
+        content += _truncate_notice(path, cap)
+        projection.issues.append(
+            ProjectionIssue(
+                projection.role,
+                category,
+                path,
+                f"exceeds context_injection.{cap_key} ({cap}); injection truncates it",
+                line,
+            )
+        )
+    return _append_block(
+        projection,
+        budget,
+        header=header or path,
+        path=path,
+        content=content,
+        size=size,
+        reason=reason,
+        category=category,
+        line=line,
+    )
+
+
+def _materialize_directory(
+    repo_root: Path,
+    entry: ContextEntry,
+    task_dir: Path,
+    limits: dict[str, int],
+    budget: _Budget,
+    projection: ContextProjection,
+) -> list[str]:
+    target = resolve_context_entry_path(entry.path, repo_root, task_dir)
+    if target is None or not _is_allowed_path(repo_root, target) or not target.is_dir():
+        reason = "directory is missing, unreadable, or escapes the allowed project roots"
+        projection.issues.append(
+            ProjectionIssue(projection.role, "directory", entry.path, reason, entry.line)
+        )
+        return []
+    try:
+        files = sorted(
+            child for child in target.iterdir() if child.suffix == ".md" and child.is_file()
+        )
+    except OSError:
+        projection.issues.append(
+            ProjectionIssue(projection.role, "directory", entry.path, "directory is unreadable", entry.line)
+        )
+        return []
+
+    if not files:
+        projection.issues.append(
+            ProjectionIssue(
+                projection.role, "directory", entry.path, "directory has no direct Markdown files to inject", entry.line
+            )
+        )
+        return []
+    if len(files) > DIRECTORY_MAX_FILES:
+        projection.issues.append(
+            ProjectionIssue(
+                projection.role,
+                "directory",
+                entry.path,
+                f"directory contains {len(files)} Markdown files; injection selects only the first {DIRECTORY_MAX_FILES}",
+                entry.line,
+            )
+        )
+
+    blocks: list[str] = []
+    for child in files[:DIRECTORY_MAX_FILES]:
+        display = f"{entry.path.rstrip('/')}/{child.name}"
+        block = _materialize_file(
+            repo_root,
+            child,
+            display,
+            entry.reason,
+            limits,
+            budget,
+            projection,
+            category="file",
+            line=entry.line,
+        )
+        if block:
+            blocks.append(block)
+    return blocks
+
+
+def project_agent_context(
+    repo_root: Path, task_dir: Path, agent_type: str, limits: dict[str, int]
+) -> ContextProjection:
+    """Render the exact Python Hook context and record every loss of body text."""
+    jsonl_name = f"{agent_type}.jsonl"
+    entries, manifest_problems, manifest_exists = parse_context_manifest(task_dir / jsonl_name)
+    projection = _new_projection(agent_type, entries, manifest_problems, manifest_exists)
+    budget = _Budget(limits["max_total_bytes"])
+    blocks: list[str] = []
+
+    for entry in entries:
+        if entry.entry_type == "directory":
+            blocks.extend(_materialize_directory(repo_root, entry, task_dir, limits, budget, projection))
+            continue
+        target = resolve_context_entry_path(entry.path, repo_root, task_dir)
+        block = _materialize_file(
+            repo_root,
+            target,
+            entry.path,
+            entry.reason,
+            limits,
+            budget,
+            projection,
+            category="file",
+            line=entry.line,
+        )
+        if block:
+            blocks.append(block)
+
+    if blocks:
+        context_parts = ["\n\n".join(blocks)]
+    else:
+        context_parts = [
+            f"[Trellis] {_display_task_path(repo_root, task_dir, jsonl_name)} has no curated entries, so no spec/research "
+            "context was injected. Before working, read the guidelines relevant to the code you will touch under "
+            ".trellis/spec/, and treat the task artifacts below as the only prepared context."
+        ]
+
+    artifact_specs = (
+        ("prd.md", "Requirements", "Requirements document"),
+        ("design.md", "Technical Design", "Technical design document"),
+        ("implement.md", "Execution Plan", "Execution plan document"),
+    )
+    for file_name, label, reason in artifact_specs:
+        target = task_dir / file_name
+        if not target.exists():
+            continue
+        display = _display_task_path(repo_root, task_dir, file_name)
+        block = _materialize_file(
+            repo_root,
+            target,
+            display,
+            reason,
+            limits,
+            budget,
+            projection,
+            category="artifact",
+            line=None,
+            header=f"{display} ({label})",
+            cap_key="max_artifact_bytes",
+        )
+        if block:
+            context_parts.append(block)
+
+    projection.text = "\n\n".join(context_parts)
+    return projection