@brandry/claude-jsonl-compressor 1.0.0 → 1.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1,78 +1,78 @@
1
- # Codex Offline Compression Summary
2
-
3
- This is a candidate Claude Code JSONL compact summary generated outside Claude. It summarizes only older records from the selected active chain. Rewound, inactive, and unattributed records are excluded.
4
-
5
- ## 1. Scope
6
-
7
- - Source file label: {input_path}
8
- - Original record count: {total_records}
9
- - Active-chain records selected for summary: {omitted_record_count} records before the preserved active-chain window
10
- - Recent raw active-chain records preserved: {recent_record_count} records
11
- - Physical line window of preserved records: {recent_start} to {recent_end}
12
- - Time span summarized: {first_ts} to {last_omitted_ts}
13
- - Time span preserved: {first_kept_ts} to {last_kept_ts}
14
- - Session distribution: {session_counts}
15
- - Common working directories: {cwd_counts}
16
- - Claude Code versions: {version_counts}
17
-
18
- ## 2. Structural Overview
19
-
20
- - Record types: {type_counts}
21
- - System subtypes: {subtype_counts}
22
- - Tool invocation overview: {tool_counts}
23
- - Attachment / hook overview: {attachment_counts}
24
- - File history snapshots: {file_history_count}
25
- - Existing compact summaries summarized into this layer: {existing_compact_count}
26
- - Human user prompts: {human_user_count}
27
- - Tool-result user records: {tool_result_user_count}
28
-
29
- ## 3. Long-Term Memory Ledger
30
-
31
- {long_term_memory}
32
-
33
- ## 4. Early / Middle Summary
34
-
35
- ### 4.1 Early
36
- {early_summary}
37
-
38
- ### 4.2 Middle
39
- {middle_summary}
40
-
41
- ## 5. Assistant Behavior and Evidence
42
-
43
- ### 5.1 Assistant research decisions and rationales
44
- {assistant_decision_items}
45
-
46
- ### 5.2 Key assistant outputs
47
- {assistant_items}
48
-
49
- ### 5.3 Key paths and filenames
50
- {path_counts}
51
-
52
- ### 5.4 Errors and anomalies
53
- {error_section}
54
-
55
- ## 6. Existing Compact Records
56
-
57
- ### 6.1 compact_boundary layer
58
- {compact_boundary_items}
59
-
60
- ### 6.2 isCompactSummary layer
61
- {compact_items}
62
-
63
- ### 6.3 Repeated compression policy
64
- - Treat previous compact layers as prior memory, not as live stacked context.
65
- - Fold still-relevant facts into the current compact summary.
66
- - Keep provenance, hashes, file labels, and line ranges in metadata or sidecars.
67
- - If prior layers themselves are too large, switch to a long-term memory ledger.
68
-
69
- ## 7. Recent Raw Preservation
70
-
71
- {recent_preservation_notes}
72
-
73
- ## 8. Important Reminders
74
-
75
- - Exact wording lives in the source JSONL or external archives.
76
- - This file is a candidate transcript rewrite, not a proof of Claude runtime behavior.
77
- - For humanities, law, art, strategy, planning, history, feasibility, and document-research sessions, preserve user goals, reasons, rejected alternatives, version changes, provenance, unresolved questions, and risk judgments.
78
- - Project-specific facts must come from the JSONL or an explicit handoff summary, not from hardcoded skill memory.
1
+ # Codex Offline Compression Summary
2
+
3
+ This is a candidate Claude Code JSONL compact summary generated outside Claude. It summarizes only older records from the selected active chain. Rewound, inactive, and unattributed records are excluded.
4
+
5
+ ## 1. Scope
6
+
7
+ - Source file label: {input_path}
8
+ - Original record count: {total_records}
9
+ - Active-chain records selected for summary: {omitted_record_count} records before the preserved active-chain window
10
+ - Recent raw active-chain records preserved: {recent_record_count} records
11
+ - Physical line window of preserved records: {recent_start} to {recent_end}
12
+ - Time span summarized: {first_ts} to {last_omitted_ts}
13
+ - Time span preserved: {first_kept_ts} to {last_kept_ts}
14
+ - Session distribution: {session_counts}
15
+ - Common working directories: {cwd_counts}
16
+ - Claude Code versions: {version_counts}
17
+
18
+ ## 2. Structural Overview
19
+
20
+ - Record types: {type_counts}
21
+ - System subtypes: {subtype_counts}
22
+ - Tool invocation overview: {tool_counts}
23
+ - Attachment / hook overview: {attachment_counts}
24
+ - File history snapshots: {file_history_count}
25
+ - Existing compact summaries summarized into this layer: {existing_compact_count}
26
+ - Human user prompts: {human_user_count}
27
+ - Tool-result user records: {tool_result_user_count}
28
+
29
+ ## 3. Long-Term Memory Ledger
30
+
31
+ {long_term_memory}
32
+
33
+ ## 4. Early / Middle Summary
34
+
35
+ ### 4.1 Early
36
+ {early_summary}
37
+
38
+ ### 4.2 Middle
39
+ {middle_summary}
40
+
41
+ ## 5. Assistant Behavior and Evidence
42
+
43
+ ### 5.1 Assistant research decisions and rationales
44
+ {assistant_decision_items}
45
+
46
+ ### 5.2 Key assistant outputs
47
+ {assistant_items}
48
+
49
+ ### 5.3 Key paths and filenames
50
+ {path_counts}
51
+
52
+ ### 5.4 Errors and anomalies
53
+ {error_section}
54
+
55
+ ## 6. Existing Compact Records
56
+
57
+ ### 6.1 compact_boundary layer
58
+ {compact_boundary_items}
59
+
60
+ ### 6.2 isCompactSummary layer
61
+ {compact_items}
62
+
63
+ ### 6.3 Repeated compression policy
64
+ - Treat previous compact layers as prior memory, not as live stacked context.
65
+ - Fold still-relevant facts into the current compact summary.
66
+ - Keep provenance, hashes, file labels, and line ranges in metadata or sidecars.
67
+ - If prior layers themselves are too large, switch to a long-term memory ledger.
68
+
69
+ ## 7. Recent Raw Preservation
70
+
71
+ {recent_preservation_notes}
72
+
73
+ ## 8. Important Reminders
74
+
75
+ - Exact wording lives in the source JSONL or external archives.
76
+ - This file is a candidate transcript rewrite, not a proof of Claude runtime behavior.
77
+ - For humanities, law, art, strategy, planning, history, feasibility, and document-research sessions, preserve user goals, reasons, rejected alternatives, version changes, provenance, unresolved questions, and risk judgments.
78
+ - Project-specific facts must come from the JSONL or an explicit handoff summary, not from hardcoded skill memory.