v1.5.2

What's Changed

• Remove per-session USD budget caps by @calebevans in #149
• Make the auto-compaction limit configurable by @calebevans in #151
• fix(db): quote every non-bareword FTS5 token, not just a known list by @calebevans in #155
• fix: make planners emit a plan before the turn limit and recover when they don't by @calebevans in #156
• fix: build ewfmount against FUSE 2 so E01 mounts work in the image by @calebevans in #159
• fix: robust structured-output parsing for leaked tool-call markup by @calebevans in #157
• fix(report): truthful, grammatical evidence-file hash sentence by @calebevans in #163
• fix(extract): skip run_mft_parser on non-NTFS volumes by @calebevans in #165
• fix(session): stop Claude Code auto-titling every agent session by @calebevans in #166
• fix: enforce role tool lists, verify MCP tools at session start, fail idle executors by @calebevans in #167
• Ship botocore[crt] so mounted aws login sessions work in the container by @calebevans in #169
• fix: validate batch task arguments against tool signatures; correct args on retry by @calebevans in #171
• docs: note BuildKit requirement for building the image by @calebevans in #174
• fix: trim leaked tool-call markup from the edges of prose text blocks by @calebevans in #177
• fix: reconcile planner prompts with executor role allowlists; drop off-role plan tasks by @calebevans in #178
• feat: render planner tool lists from executor allowlists by @calebevans in #179
• fix: mount disk images over FUSE so the unprivileged container user can use them by @calebevans in #180
• fix: trim blank edge lines so a leaked marker behind them is removed by @calebevans in #183
• fix: validate bulk_extractor scanner names before invoking the binary by @calebevans in #184
• fix: skip the YARA rule pull when the signature-base clone is read-only by @calebevans in #186
• fix: pass every EWF segment to xmount so split E01 images mount by @calebevans in #190
• fix: fill implicit case_id in batch and parallel dispatch instead of rejecting by @calebevans in #188
• fix: tell Claude Code the real context window and output cap for gateway models by @calebevans in #192
• fix: forward reasoning_effort raw so proxy models actually think by @calebevans in #194
• fix: remediate split-phase gate gaps, continue after max_turns, warn on missing event_time at submit by @calebevans in #196
• fix: count only extraction calls in the idle-executor guard by @calebevans in #198
• feat: log thinking blocks and silent assistant turns by @calebevans in #201
• fix: exempt cheap control/query tools from the resource gate and tool limiter by @calebevans in #202
• feat: per-model context/output/reasoning overrides for proxy models by @calebevans in #204
• fix: route unmapped Bedrock models through LiteLLM's explicit converse path by @calebevans in #206
• fix: allow tools for Bedrock models LiteLLM's map does not know by @calebevans in #208
• fix: invoke log2timeline with --storage_file and never prompt on stdin by @calebevans in #210
• fix: gate the report phase on the report file, not on finalize_report being called by @calebevans in #212
• fix: disable every Claude Code built-in tool in agent sessions by @calebevans in #215
• fix: ignore analyst completion messages and accept empty follow-up plans by @calebevans in #216
• fix: run the analyst when a split-phase planner produces no plan by @calebevans in #218
• fix: make deduplicate_findings lossless and guard against over-merging by @calebevans in #220
• fix: measure citation coverage over distinct evidence sources and make it advisory by @calebevans in #222
• feat: detect extension/content mismatches (masqueraded files) on disk images by @calebevans in #224
• feat: read optical media (UDF/ISO 9660) images, deleted files included by @calebevans in #226
• fix: scope tsk.partitions lookups per image; scan every partition in detect_masquerading by @calebevans in #228
• chore: bump version to 1.5.2 by @calebevans in #229

Full Changelog: v1.5.1...v1.5.2