csess 0.8.1

Fast lister for Claude Code sessions in a folder and its subprojects
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141

# Changelog

All notable changes to this project are documented here. The format is based on
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and this project adheres
to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [0.8.1] - 2026-07-10

### Fixed
- No longer panics with `failed printing to stdout: Broken pipe` when the reader
  hangs up early (`csess --show … | head`, quitting `less`, or a lazy-load UI
  that stops reading mid-transcript). Output now exits cleanly (status 0) on a
  broken pipe instead of aborting with a Rust backtrace. Pre-existing since the
  first release; affects all subcommands.

## [0.8.0] - 2026-07-10

### Added
- `--show --json` (and `--grep --json`) now pass through more raw per-entry
  fields on each message so a consumer can classify plumbing/subagent turns
  itself instead of guessing from text prefixes: `is_meta` (`isMeta`),
  `user_type` (`userType`, e.g. `"external"`), and `entry_type` (raw entry
  `type`). All verbatim mirrors, no classification.

### Changed
- **Breaking (JSON):** `schema_version` bumped to `3`. Two existing fields were
  renamed to match their raw JSONL keys: `sidechain` → `is_sidechain` and
  `tool_result` → `tool_use_result`. `is_sidechain`/`is_meta` are now always
  present booleans (previously `sidechain` was omitted when false).
- `isMeta` turns (skill-load / image-read / attachment echoes, `/loop`
  heartbeats, and the "Continue from where you left off." resume line) are no
  longer dropped from `--show` or `message_count` — they surface tagged
  `is_meta:true` for the consumer to suppress. The only entry still dropped is
  the `model:"<synthetic>"` "No response requested." reply, identified by that
  model sentinel rather than by matching English text (verified language-
  independent across 400 local sessions). `message_count` grows for sessions
  that contain meta turns.

## [0.7.0] - 2026-07-10

### Added
- `--show --json` now carries the `toolUseResult` sidecar Claude Code logs
  alongside each tool result, passed through verbatim on the message's
  `tool_result` field. This exposes `structuredPatch` (Edit/Write diff hunks —
  what the desktop/mobile UI renders as a colored diff with `+N -M` counts) and
  `stdout`/`stderr`/`interrupted` (Bash), so a UI consuming csess can render
  tool calls, file edits, and command output the way the official app does
  rather than only flattened text. Absent on messages without a sidecar.

### Changed
- **Breaking (JSON):** `schema_version` bumped to `2`. The only change is the
  additive `tool_result` field above; existing fields are untouched, so v1
  consumers keep working, but the version signals the new field is available.

## [0.6.0] - 2026-07-07

### Added
- `--after <id>` pages the session list — a list-level cursor mirroring
  `--before` on `--show`. It drops every session up to and including the given
  id in the current sort order, then applies `-n`, so `csess -n 12` then
  `csess -n 12 --after <last-id>` walks the list a page at a time. Keyed by id
  (full or short) rather than a numeric offset, so a session bumping to the top
  between pages never duplicates or skips a row; an unknown id is a no-op.
  Trims output only — every file is still parsed, so accuracy is unchanged.

## [0.5.0] - 2026-06-27

### Added
- `--role user|assistant` filters a transcript to one side. Works with `--show`
  and with `--grep`.
- `--grep <text>` searches message text (case-insensitive substring over the
  flattened, human-readable content — tool calls included, matching what you
  see). With `--show` it filters that one transcript; without `--show` it scans
  every in-scope session in parallel and prints each match's session plus its
  matching messages as one-line snippets. Honours `-g`/`--period`/`--since` and
  `--json` (output shape `{ "schema_version": 1, "matches": [...] }`).

## [0.4.0] - 2026-06-25

### Added
- `schema_version` field on all `--json` output (currently `1`). Downstream
  tools can pin to it; field renames/removals/retypes bump it, additive fields
  don't.

### Changed
- **Breaking:** the list `--json` output is now an object
  `{ "schema_version": 1, "sessions": [...] }` instead of a bare array. Consumers
  that parsed the top-level array must read the `sessions` key. The `--show
  --json` transcript is unchanged apart from gaining `schema_version` (additive).

## [0.3.0] - 2026-06-21

### Added
- `--show --before <uuid>` pages a transcript backwards: it returns only the
  messages older than the given message uuid, then `-n` keeps the last N of
  those. A uuid cursor (not an offset) for scroll-up / lazy-load chat UIs —
  immune to index drift when the active session appends new turns.

### Changed
- `--show` and `message_count` now skip the synthetic resume entries Claude Code
  writes to its own log (the `isMeta` "Continue from where you left off." and the
  `model: "<synthetic>"` "No response requested." pairs). Transcripts and counts
  now match what the Claude Code UI displays instead of over-reporting.

## [0.2.1] - 2026-06-21

### Added
- `csess --version` now reports the Claude Code version this release was tested
  against (e.g. `0.2.1 (verified with Claude Code 2.1.185)`), so `.jsonl` format
  drift in a future Claude Code release can be traced to a known-good baseline.

## [0.2.0] - 2026-06-21

### Added
- `--show <id-or-name>` prints a single session's full transcript. Matches by
  full id, short id, or a name substring; ambiguous matches list candidates and
  exit non-zero. Use `-g` to find a session in any project.
- `--show --json` emits a structured transcript (session metadata +
  `messages[]` with `role`, `timestamp`, `uuid`, `sidechain`, and the raw
  Anthropic `content` blocks) for re-rendering a conversation in another UI.
  Plain `--show` prints a human-readable view.
- `-n/--limit` applied to `--show` keeps only the last N messages (tail).
- SIZE column in the listing table (size was already in `--json`).

## [0.1.1] - 2026-06-19

### Fixed
- Session name now reads the `ai-title` line Claude Code writes, matching the
  native `/resume` picker. Name precedence is `aiTitle` → `summary` →
  first real prompt.
- Skip `<ide_...>` selections and `Base directory for this skill:` injections
  when falling back to the first prompt for a name.

## [0.1.0] - 2026-06-16

### Added
- List Claude Code sessions for a folder and its subfolder projects.
- Recursive discovery with loose encoded pre-filter + strict `cwd` verification.
- Fuzzy search (`--search`), time filters (`--since`/`--until`/`--period`),
  sorting (`--sort`), `--limit`, `--global`.
- Table (default) and `--json` output.