Kiro CLI 1.0.288 sessions not discovered — transcripts moved to ~/.kiro/sessions/<workspace-hash>/sess_<id>/messages.jsonl, not ~/.kiro/sessions/cli/*.jsonl

[lachlansimpson]

agentsview version

agentsview v0.41.1 (commit a902515, built 2026-08-18T13:35:26Z)

Install method

install.sh / install.ps1

OS / platform

Ubuntu 24.04.4 LTS (x86_64)

Which agent and version

Kiro CLI 1.0.288 (ec90738d3f0da2389adcb94a6661b5abf1fc20e8, x64)

Which model(s)

qdev::auto — it's on reasoningModelId in assistant records; session.json.modelId is null

What happened, and what did you expect

The kiro provider only discovers the flat legacy layout ~/.kiro/sessions/cli/<uuid>.jsonl. Kiro CLI now writes transcripts per workspace instead:

~/.kiro/sessions/<workspace-hash>/sess_<uuid>/messages.jsonl

So no Kiro session has been ingested on this machine since 2026-08-13, although I use Kiro daily and today's sessions are on disk. Expected: sessions created after the layout change show up like any other agent's.

The daemon is otherwise healthy — it incrementally syncs Claude Code sessions in the same watch cycle, and no error is logged. The only hint is the discovery count, which looks plausible (~/.agentsview/debug.log):

2026/08/24 10:39:27 discovered 297 files (93 claude, 1 codex, 159 copilot, 0 gemini,
  0 cursor, 0 amp, 0 zencoder, 0 iflow, 16 vscode-copilot, 0 visualstudio-copilot,
  0 pi, 0 omp, 24 kiro, 0 zed, 0 vibe) in 93ms

That 24 kiro is exactly my 23 stale legacy .jsonl files plus ~/.local/share/kiro-cli/data.sqlite3. The 32 current transcripts under the new layout are not in the set. Newest kiro rows in ~/.agentsview/sessions.db confirm it:

kiro:e8e37560-c9c7-44fa-9412-d5ebe79ade3d  ended 2026-08-13T06:08:44Z  176 msgs
kiro:ea09aafc-0c15-4e2d-8514-a0df1492b28a  ended 2026-08-13T05:46:09Z  110 msgs

Two things make this harder to work around than it looks:

• ~/.local/share/kiro-cli/data.sqlite3 is already known to the binary (kiroDBPathForEvent, kiroDBUnderRoot), but on 1.0.288 both conversations and conversations_v2 are empty (0 rows). Transcripts are filesystem-only.
• ~/.kiro/sessions/cli/ still gets a sess_<uuid>.history file per session — prompt-input history, no transcript — so the legacy directory keeps a fresh mtime while holding nothing ingestible.
• There's no KIRO_SESSIONS_DIR equivalent to CODEX_SESSIONS_DIR / TRAEX_SESSIONS_DIR / VSCODE_COPILOT_DIR, so there's no user-side override either.

Suggested fix, alongside kiroLegacyPathUnderRoot: also walk ~/.kiro/sessions/*/sess_*/messages.jsonl; dispatch the parser on record shape ({kind,version,data} → legacy, {id,timestamp,payload} → new); take the session id from the sess_ directory name and title/timestamps/project from the sibling session.json (workspacePaths[0] is more reliable than reversing the workspace hash); skip sessions/cli/sess_*.history. Legacy sessions keep parsing unchanged, so this can be purely additive. A KIRO_SESSIONS_DIR override would be a bonus.

The kiro provider only discovers the flat legacy layout ~/.kiro/sessions/cli/<uuid>.jsonl. Kiro CLI now writes transcripts per workspace instead:

~/.kiro/sessions/<workspace-hash>/sess_<uuid>/messages.jsonl

So no Kiro session has been ingested on this machine since 2026-08-13, although I use Kiro daily and today's sessions are on disk. Expected: sessions created after the layout change show up like any other agent's.

The daemon is otherwise healthy — it incrementally syncs Claude Code sessions in the same watch cycle, and no error is logged. The only hint is the discovery count, which looks plausible (~/.agentsview/debug.log):

2026/08/24 10:39:27 discovered 297 files (93 claude, 1 codex, 159 copilot, 0 gemini,
    0 cursor, 0 amp, 0 zencoder, 0 iflow, 16 vscode-copilot, 0 visualstudio-copilot,
    0 pi, 0 omp, 24 kiro, 0 zed, 0 vibe) in 93ms

That 24 kiro is exactly my 23 stale legacy .jsonl files plus ~/.local/share/kiro-cli/data.sqlite3. The 32 current transcripts under the new layout are not in the set. Newest kiro rows in ~/.agentsview/sessions.db confirm it:

  kiro:e8e37560-c9c7-44fa-9412-d5ebe79ade3d  ended 2026-08-13T06:08:44Z  176 msgs
  kiro:ea09aafc-0c15-4e2d-8514-a0df1492b28a  ended 2026-08-13T05:46:09Z  110 msgs

Two things make this harder to work around than it looks:

• ~/.local/share/kiro-cli/data.sqlite3 is already known to the binary (kiroDBPathForEvent, kiroDBUnderRoot), but on 1.0.288 both conversations and conversations_v2 are empty (0 rows). Transcripts are filesystem-only.
• ~/.kiro/sessions/cli/ still gets a sess_<uuid>.history file per session — prompt-input history, no transcript — so the legacy directory keeps a fresh mtime while holding nothing ingestible.
• KIRO_SESSIONS_DIR / kiro_dirs exist, but can't express the new layout: legacy discovery only matches .jsonl files directly in the root, so pointing either at ~/.kiro/sessions still finds nothing two levels down.

Suggested fix, alongside kiroLegacyPathUnderRoot: also walk ~/.kiro/sessions/*/sess_*/messages.jsonl; dispatch the parser on record shape ({kind,version,data} → legacy, {id,timestamp,payload} → new); take the session id from the sess_<uuid> directory name and title/timestamps/project from the sibling session.json (workspacePaths[0] is more reliable than reversing the workspace hash); skip sessions/cli/sess_*.history. Legacy sessions keep parsing unchanged, so this can be purely additive. A KIRO_SESSIONS_DIR override would be a bonus.

Sample session file or snippet

Layout comparison
=================

LEGACY (last written 2026-08-13 on my machine)
  ~/.kiro/sessions/cli/<uuid>.jsonl        <- transcript
  ~/.kiro/sessions/cli/<uuid>.json
  ~/.kiro/sessions/cli/<uuid>.history

CURRENT (Kiro CLI 1.0.288)
  ~/.kiro/sessions/<workspace-hash>/sess_<uuid>/messages.jsonl   <- transcript
  ~/.kiro/sessions/<workspace-hash>/sess_<uuid>/session.json     <- metadata
  ~/.kiro/sessions/<workspace-hash>/sess_<uuid>/snapshots/<id>/  <- file snapshots
  ~/.kiro/sessions/<workspace-hash>/sess_<uuid>/publish.cursor
  ~/.kiro/sessions/cli/sess_<uuid>.history                       <- prompt history only


Legacy record envelope: {kind, version, data}
=============================================
{"kind": "Prompt",           "version": "v1", "data": {...}}
{"kind": "AssistantMessage", "version": "v1", "data": {...}}
{"kind": "ToolResults",      "version": "v1", "data": {...}}


New record envelope: {id, timestamp, payload}, discriminator on payload.type
===========================================================================
(content values redacted; key structure intact)

{"id":"<redacted>","timestamp":"<redacted>","payload":{"type":"user","content":"<redacted>","documents":[],"images":[],"_meta":{...}}}
{"id":"<redacted>","timestamp":"<redacted>","payload":{"type":"assistant","content":"<redacted>","executionId":"<redacted>","operationType":"<redacted>","reasoningModelId":"qdev::auto","reasoningSignature":"<redacted>","_meta":{...}}}
{"id":"<redacted>","timestamp":"<redacted>","payload":{"type":"tool_call","toolName":"<redacted>","toolCallId":"<redacted>","args":"<redacted>","actionType":"<redacted>","kind":"<redacted>","status":"<redacted>","title":"<redacted>","executionId":"<redacted>","_meta":{...}}}
{"id":"<redacted>","timestamp":"<redacted>","payload":{"type":"tool_result","toolCallId":"<redacted>","content":"<redacted>","success":true,"executionId":"<redacted>","_meta":{...}}}


payload.type histogram, one 258 KB session (314 records)
========================================================
   77  assistant
   59  session_metadata          (key is always "contextUsage" in this session)
   57  tool_call
   57  tool_result
   14  pending_interaction
   14  interaction_resolved
   14  ContextualHookInvoked
    4  user
    4  turn_start
    4  usage_summary
    4  session_event
    4  turn_end
    1  steering_inclusion
    1  session_start

turn_start / turn_end should make turn boundaries cheaper to derive than in the legacy
format. Caveat on usage: usage_summary carries no token counts. Shape is
{type, executionId, status, elapsedTime, requestIds[], promptTurnSummaries[]}, and each
promptTurnSummaries entry is {unit, unitPlural, usage: <float>, usedTools[]} — i.e. Kiro
credits, not tokens.


session.json (values redacted, keys intact)
===========================================
{
  "id": "<redacted>",
  "schemaVersion": "1.0.0",
  "dataModelVersion": 1,
  "title": "<redacted>",
  "description": "<redacted>",
  "status": "idle",
  "agentMode": "vibe",
  "modelId": null,
  "createdAt": "2026-08-23T23:54:14.859Z",
  "lastModifiedAt": "2026-08-24T00:51:42.067Z",
  "workspacePaths": ["<redacted absolute path to project>"],
  "autopilot": ..., "ftaEnabled": ..., "semanticReviewEnabled": ...,
  "specPlanEnabled": ..., "specSkipClarificationEnabled": ...,
  "specWorkflow": ..., "workflowsEnabled": ..., "_meta": {...}
}

modelId is null; the live model appears as reasoningModelId on assistant records.
workspacePaths[0] is the project path. (~/.kiro/workspace-roots/<hash>/.trust-migration.json
also has a "root" field, but only for workspaces that went through trust migration, so it
is not a complete hash -> path map.)

Steps to reproduce

1. Install Kiro CLI 1.0.288 and run a session in any workspace.
2. Confirm the transcript exists under the new layout:
   find ~/.kiro/sessions -mindepth 3 -name messages.jsonl -newermt today
3. Confirm the legacy directory got only a prompt-history file, no transcript:
   ls -la ~/.kiro/sessions/cli/sess_*.history
4. Trigger a sync (agentsview sync, or wait for the watcher) and check the discovery line in ~/.agentsview/debug.log — the kiro count does not include the new sessions.
5. Query the DB — nothing after the layout change:
   sqlite3 ~/.agentsview/sessions.db "select id, ended_at from sessions where id like 'kiro%' order by ended_at desc limit 5;"

Checklist

• I searched existing issues
• I removed secrets and private data from any attached session files