Digital Foundry claims Actualis can locally reconstruct and expose what coding agents did on a developer’s machine, making agent actions and outcomes more practically auditable.

state: expiredheat: lowuncertainty: highknownscott: mediumcoding-agents agent-observabilityDigital FoundryActualis

Surfaced 2026-08-29T20:27:27Z — priced heat=high at reprice: The case now exposes a substrate-level limitation: local transcript reconstruction is not durable auditing if Claude Code silently deletes its source records after 30 days. This does not validate or invalidate Actualis’s reconstruction accuracy, but it creates an actionable retention risk for observability and receipt systems built on those files.

What is this?

Actualis is presented as local software that reconstructs and exposes what coding agents did on a developer’s machine, with an original repository apparently named `agentfleet`. The broader snippets establish the problem—coding agents can execute commands, access files, and make network connections without continuous supervision—and identify tracing as a way to improve visibility and trust. However, the supplied results do not directly document Actualis, its mechanism, or its provenance, and they reference multiple unrelated entities named Digital Foundry, so the developer and attribution remain unverified.

Why it matters to Scott

This repeats Scott’s established Agent Observability and Agent Receipts position and directly resembles his Search Conversations session-reconstruction project. The radar already tracks substantially similar developments in Rungraph, Hindcast, ProofRun, and Numbat; with Actualis’s mechanism and provenance still unverified, the case adds no demonstrated capability or claim that would change what Scott builds or argues.

ip:concept.agent-observabilityip:concept.agent-receiptsdev:project.search-conversationsradar:rungraph-claude-code-session-replayradar:hindcast-claude-code-session-recoveryradar:proofrun-local-agent-verification-receiptsradar:numbat-agent-endpoint-visibilityradar:concept.agent-observability

queries asked of Scott's wikis

coding-agent action logs and audit trails
local agent observability and session reconstruction
coding-agent harness provenance and reproducibility
auditing agent file, shell, and network actions
agent accountability beyond tracing
developer-machine security for autonomous coding agents

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (8) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn	Actualis – read what your coding agent did on your machine	actualis	3	0
🟧 echo.github ⭐	This is the repository’s root commit, with no parent or upstream source. Its original README introduced `agentfleet` as software that “reads	Digital Foundry Solutions	—	—
🟧 hn	Show HN: Slaunt – Know what your agents are doing	Kerode0	1	0
🟠 reddit	How do you actually know what your agent did while you weren't watching? ClaudeAI	Independent_Bag_2904	1	14
🟠 reddit	Six subagents on three different models in one turn. I built something to actually watch that happen while it runs. ClaudeAI	duqaxxx	3	8
🟧 hn	Show HN: Seedeep – I couldn't see what Claude Code was doing, so I drew it	duqaxxx	1	0
🟠 reddit	PSA: Claude Code silently deletes local history older than 30 days ClaudeAI	fobax	117	61
🟠 reddit	We built a retrospective reader for Claude Code using Claude ClaudeAI	rohynal	1	2

Interpretation history

2026-08-30T09:28:25Z
Repeated comment refreshes only elaborate the already-surfaced retention mitigations and storage tradeoffs; they still provide no inspection, adoption evidence, or validation of Actualis itself. The broader reconstruction pattern is established enough to retain, but this Actualis-centered episode has faded without an expected confirming event.
2026-08-30T08:24:22Z
The refreshed discussion remains mitigation and storage commentary around the already-surfaced retention risk, with no new validation of Actualis or durable audit guarantees. The broader local reconstruction pattern is corroborated, but this episode has no fresh momentum.
2026-08-30T06:32:14Z
Refreshed comments add operational detail about transcript storage costs and preservation practices, but no independent confirmation or validation of Actualis’s reconstruction quality. The category remains corroborated while durable auditworthiness and Actualis’s specific claims stay unsettled.
2026-08-30T04:28:02Z
The refreshed retention discussion adds no new confirmation, failure evidence, or validation of Actualis’s reconstruction quality. Local session reconstruction remains a corroborated product pattern, but durable auditworthiness and Actualis’s specific claims remain unsettled.
2026-08-30T03:23:17Z
A third self-reported reader reinforces local session reconstruction as a recurring coding-agent observability pattern, but shows no adoption, independent testing, or durable audit guarantees. Actualis’s own accuracy and provenance remain unvalidated, so this adds category density rather than momentum.
2026-08-30T03:22:34Z
evidence attached: reddit.post.1w25pyv — This open-source retrospective reader independently corroborates the value of reconstructing and governing coding-agent execution behavior.
2026-08-29T23:24:03Z
The refreshed discussion adds no independent test, first-party confirmation, or evidence about Actualis’s reconstruction quality; it remains repetition of the already-surfaced retention risk and preservation workarounds. The broader local session-reconstruction pattern is corroborated, but this episode has no new momentum.
2026-08-29T22:39:32Z
Refreshed comments repeat preservation workarounds and the known risk of trusting agent-authored reports; they add no validation of Actualis, Seedeep, or the reported cleanup behavior. The broader local-reconstruction pattern remains corroborated, but this episode has no fresh momentum and can cool.
2026-08-29T21:29:39Z
Refreshed comments add concrete preservation workarounds—external SQLite logging, pre-startup backup, and controlled retention testing—but no independent validation of the cleanup behavior or Actualis’s audit quality. The operational risk was already surfaced, so this is mitigation detail rather than a new escalation.
2026-08-29T20:27:27Z
The case now exposes a substrate-level limitation: local transcript reconstruction is not durable auditing if Claude Code silently deletes its source records after 30 days. This does not validate or invalidate Actualis’s reconstruction accuracy, but it creates an actionable retention risk for observability and receipt systems built on those files.
2026-08-29T20:23:31Z
evidence attached: reddit.post.1w1wc0u — An independent user report of silent, permanent Claude Code transcript deletion materially undermines the practical auditability of coding-agent activity.
2026-08-29T16:29:58Z
Seedeep provides a second independent implementation of local coding-agent session reconstruction, corroborating the broader product pattern without validating Actualis’s accuracy, provenance, or auditworthiness. The HN appearance is duplicate distribution rather than a further capability or adoption signal, so the case can cool.
2026-08-29T16:23:47Z
evidence attached: hn.story.49490770 — Seedeep is an independent coding-agent observability artifact that bears directly on whether Claude Code actions can be made visible and auditable.
2026-08-29T15:34:56Z
The refreshed discussion sharpens the distinction between reconstructing session activity and producing trustworthy audit receipts: agent-authored reports may themselves be unreliable. It adds a useful failure mode but no technical inspection or independent validation of Actualis or Seedeep.
2026-08-29T14:25:20Z
Seedeep independently implements live reconstruction from Claude Code’s local session files, moving the approach from an isolated Actualis claim toward a recurring product pattern. It supports technical plausibility but still does not validate Actualis’s provenance, reconstruction accuracy, or auditing quality.
2026-08-29T14:23:35Z
evidence attached: reddit.post.1w1njpj — The released session-observability tool independently supports the case that coding-agent actions and subagent behavior can be made practically auditable.
2026-08-28T23:25:22Z
Refreshed comments remain routine workarounds and expressions of the known observability gap; they add no independent use, technical inspection, or validation of Actualis. The case still represents an unverified implementation in an already-established product category.
2026-08-28T18:41:36Z
Refreshed discussion adds an adjacent open-source intent-log product and familiar user workarounds, reinforcing that agent visibility is a real product category but not validating Actualis’s reconstruction or audit quality. The case remains a thin, unverified implementation claim.
2026-08-28T17:35:29Z
The user anecdote independently reinforces the observability problem but does not validate Actualis’s reconstruction mechanism, provenance, or practical auditing quality. The case remains a thin implementation claim within an already established pattern.
2026-08-28T17:25:07Z
evidence attached: reddit.post.1w0wecx — Independent user evidence highlights the concrete observability gap that tools like Actualis aim to solve for unattended coding agents.
2026-08-27T13:36:02Z
Slaunt weakly suggests continued product interest in coding-agent visibility, but supplies no mechanism, artifact, or validation of Actualis’s reconstruction claims. The case remains an unverified implementation of an already established pattern.
2026-08-27T13:24:35Z
evidence attached: hn.story.49464222 — A second agent-observability product may materially contextualize whether practical visibility into agent actions is becoming a useful pattern, though evidence is currently thin.
2026-08-26T18:41:57Z
No new substantive evidence validates Actualis’s reconstruction or auditing claims; the lone additional comment does not change its status as an unverified implementation of an already established pattern.
2026-08-26T18:40:03Z
grounded: known/low — This repeats Scott’s established Agent Observability and Agent Receipts position and directly resembles his Search Conversations session-reconstruction project.
2026-08-26T18:37:16Z
origin walked (codex/luna, conf 0.99): anchor hn.story.49453056 -> echo.github.96cf4f469b by Digital Foundry Solutions
2026-08-26T18:36:10Z
case created — The released repository addresses the concrete and increasingly important problem of inspecting local coding-agent activity.

Decision trace

08-30 19:28expireRepeated comment refreshes only elaborate the already-surfaced retention mitigations and storage tradeoffs; they still provide no inspection, adoption evidence, or validation of Actualis itself. The b
08-30 19:28alert_silentThe new delta is repetitive discussion of a retention risk Scott was already alerted to, with no consequential fact that would make the next briefing too late.
08-30 19:28alert_routeThe new delta is repetitive discussion of a retention risk Scott was already alerted to, with no consequential fact that would make the next briefing too late.
08-30 19:21sensor_dirtycomment_update
08-30 18:24repriceThe refreshed discussion remains mitigation and storage commentary around the already-surfaced retention risk, with no new validation of Actualis or durable audit guarantees. The broader local reconst
08-30 18:24alert_silentThis is repetitive amplification of the previously alerted transcript-retention issue and adds no consequential fact that Scott needs before the next briefing.
08-30 18:24alert_routeThis is repetitive amplification of the previously alerted transcript-retention issue and adds no consequential fact that Scott needs before the next briefing.
08-30 18:21sensor_dirtycomment_update
08-30 16:32repriceRefreshed comments add operational detail about transcript storage costs and preservation practices, but no independent confirmation or validation of Actualis’s reconstruction quality. The category re
08-30 16:32alert_silentThe discussion only elaborates mitigations and storage tradeoffs around the already-alerted retention risk; it adds nothing that should change Scott’s work before the next briefing.
08-30 16:32alert_routeThe discussion only elaborates mitigations and storage tradeoffs around the already-alerted retention risk; it adds nothing that should change Scott’s work before the next briefing.
08-30 16:21sensor_dirtycomment_update
08-30 14:28repriceThe refreshed retention discussion adds no new confirmation, failure evidence, or validation of Actualis’s reconstruction quality. Local session reconstruction remains a corroborated product pattern,
08-30 14:28alert_silentThe comment refresh does not materially change the already-alerted transcript-retention risk or provide a new fact that affects Scott’s work before the next briefing.
08-30 14:28alert_routeThe comment refresh does not materially change the already-alerted transcript-retention risk or provide a new fact that affects Scott’s work before the next briefing.
08-30 14:21sensor_dirtycomment_update
08-30 13:23repriceA third self-reported reader reinforces local session reconstruction as a recurring coding-agent observability pattern, but shows no adoption, independent testing, or durable audit guarantees. Actuali
08-30 13:23alert_silentThe new reader substantially overlaps implementations already tracked and does not change the previously surfaced transcript-retention risk or provide evidence that should alter Scott’s work today.
08-30 13:23alert_routeThe new reader substantially overlaps implementations already tracked and does not change the previously surfaced transcript-retention risk or provide evidence that should alter Scott’s work today.
08-30 13:22alert_silentThis is another self-reported, open-source retrospective reader of Claude Code history, offering shell, command, and file-write summaries similar to tools already tracked in this case. It provides no
08-30 13:22alert_routeThis is another self-reported, open-source retrospective reader of Claude Code history, offering shell, command, and file-write summaries similar to tools already tracked in this case. It provides no
08-30 13:22attachThis open-source retrospective reader independently corroborates the value of reconstructing and governing coding-agent execution behavior.
08-30 13:22propose_attachThis open-source retrospective reader independently corroborates the value of reconstructing and governing coding-agent execution behavior.
08-30 12:21sensor_dirtyengagement_update
08-30 09:24repriceThe refreshed discussion adds no independent test, first-party confirmation, or evidence about Actualis’s reconstruction quality; it remains repetition of the already-surfaced retention risk and prese
08-30 09:24alert_silentThe prior heads-up already covered Claude Code’s reported 30-day transcript cleanup and the need to preserve records externally or change retention settings. A comment refresh without a changed defaul
08-30 09:24alert_routeThe prior heads-up already covered Claude Code’s reported 30-day transcript cleanup and the need to preserve records externally or change retention settings. A comment refresh without a changed defaul
08-30 09:21sensor_dirtycomment_update
08-30 08:39repriceRefreshed comments repeat preservation workarounds and the known risk of trusting agent-authored reports; they add no validation of Actualis, Seedeep, or the reported cleanup behavior. The broader loc
08-30 08:39alert_silentThe prior heads-up already covered the transcript-retention risk and reversible setting change. The new discussion is mitigation and familiar failure-mode commentary, with no changed default, first-pa
08-30 08:39alert_routeThe prior heads-up already covered the transcript-retention risk and reversible setting change. The new discussion is mitigation and familiar failure-mode commentary, with no changed default, first-pa
08-30 08:21sensor_dirtycomment_update
08-30 08:21sensor_dirtycomment_update
08-30 07:29repriceRefreshed comments add concrete preservation workarounds—external SQLite logging, pre-startup backup, and controlled retention testing—but no independent validation of the cleanup behavior or Actualis
08-30 07:29alert_silentThe prior heads-up already covered the potentially destructive 30-day startup cleanup and reversible setting change. New comments only elaborate defensive practices, with no first-party confirmation,
08-30 07:29alert_routeThe prior heads-up already covered the potentially destructive 30-day startup cleanup and reversible setting change. New comments only elaborate defensive practices, with no first-party confirmation,
08-30 07:21sensor_dirtycomment_update
08-30 06:27repriceThe case now exposes a substrate-level limitation: local transcript reconstruction is not durable auditing if Claude Code silently deletes its source records after 30 days. This does not validate or i
08-30 06:27pushpriced heat=high at reprice: The case now exposes a substrate-level limitation: local transcript reconstruction is not durable auditing if Claude Code silently deletes its source records after 30 days
08-30 06:27alert_routepriced heat=high at reprice: The case now exposes a substrate-level limitation: local transcript reconstruction is not durable auditing if Claude Code silently deletes its source records after 30 days