Changelog

All notable changes to the Hermes Brain vault template. Versions correspond to GitHub Releases.

[1.21.0] — 2026-09-18 link

Added link

• Future skill forecast widget — Scripts/skill_forecast.py analyzes vault activity trends (from Tag-Trends.log) and installed skills (from Installed-Skills-Index.md) to suggest which skills to learn/install next for maximum relevance. Maps session tags to potential skill categories via a configurable mapping, identifies installed skills that already cover trending topics, and outputs actionable suggestions with reasoning. Runs as a standalone script or can be integrated into cron workflows.
• Cross-platform script table updated — added row for skill_forecast.py.

Verified link

• skill_forecast.py syntax checks clean (python -m py_compile).
• Correctly parses 70 installed skills from Installed-Skills-Index.md.
• Correctly parses 15 unique trending tags from Tag-Trends.log (top: hermes, linux, llm, automation, obsidian, video).
• Tag-to-skill mapping covers all major trending categories.
• Output logged to Skills-Notes/Skill-Forecast.log with timestamps.

[1.20.0] — 2026-09-17 link

Added link

• Automated session tagging & topic tracking — Scripts/session_tagger.py extracts topics/themes from archived session markdown using lightweight keyword pattern matching (no external API, no heavy ML deps). Runs automatically as part of hourly_archive.py after sessions are moved into YYYY/MM/DD/ directories. Maintains:
  • Per-session tags array in Daily/manifest.jsonl
  • Global tag cloud with counts in Skills-Notes/Tag-Cloud.log
  • Daily tag trends in Skills-Notes/Tag-Trends.log
• Dashboard “Tag cloud & trends” widget — renders the tag cloud as styled pills (size/opacity scaled by frequency) with hover effects, plus a 14-day trend history from Tag-Trends.log. Degrades gracefully when source files don’t exist yet.
• Cross-platform script table updated — added row for session_tagger.py.

Verified link

• session_tagger.py syntax checks clean (python -m py_compile).
• Pattern extraction tested against seeded sample sessions — correctly identifies topics like linux, hermes, python, database, automation, devops, networking, obsidian, video.
• Tag cloud and trends logs written and formatted correctly.
• All 13 dataviewjs blocks in Dashboard.md pass node --check; markdown fences balanced.
• hourly_archive.py still byte-compiles cleanly after integration.
• Scripts/vault_audit.py reports 0 broken links.

[1.19.0] — 2026-09-17 link

Added link

• Semantic search layer — Scripts/semantic_search.py builds a local vector search index over the vault’s session and knowledge files using sentence-transformers/all-MiniLM-L6-v2 (22 MB, CPU-only, no external API). Supports full rebuild (--build), incremental update (--update), CLI search (--search), and an HTTP server (--serve) for real-time DataviewJS queries.
• Dashboard “Semantic search” widget — natural-language search box on Dashboard.md that queries the local endpoint (or falls back gracefully with clear instructions). Click results to open matching notes.
• Cross-platform script table updated — added rows for semantic_search.py, archive_now.py, enrich_session.py.

Verified link

• semantic_search.py syntax checks clean (python -m py_compile).
• Index builds incrementally (only re-embeds new/changed files via SHA256 hash tracking).
• Dashboard widget renders without errors; debounced input with loading state and clickable result links.
• All existing Dashboard DataviewJS blocks still pass node --check; markdown fences balanced.

[1.18.0] — 2026-09-17 link

Added link

• Manifest concurrency lock + instant archive trigger — Scripts/hourly_archive.py (and .ps1) now acquire an exclusive lock file (.hourly_archive.lock) to prevent concurrent execution with itself or the new manual trigger. Added Scripts/archive_now.py (and .ps1) for on-demand, instant exports (default: last 5 minutes) that share the same lock so they never race the hourly cron. Both scripts export redacted markdown + JSONL, update manifest.jsonl, and append to Token-Usage.log.
• Optional session enrichment — Scripts/enrich_session.py (and .ps1) adds a simple summary: frontmatter (first non-empty line, capped at 200 chars) to session markdown files that lack one, making sessions more glanceable without opening each file.
• Documentation updates — Daily/README.md explains the two archiving paths (hourly cron vs instant manual) and the enrichment step. INSTALL_PROMPT.md Notes section now mentions the optional instant archive and enrichment scripts.

Verified link

• All scripts (hourly_archive.py, archive_now.py, enrich_session.py) and their PowerShell twins parse cleanly (python -m py_compile, PowerShell parser).
• Throwaway-vault test: ran archive_now.py --since 5m on a seeded fixture, verified new sessions landed in correct Daily/YYYY/MM/DD/ folder, manifest.jsonl updated, token usage logged, and no duplicate entries on re‑run.
• Enrichment test: ran enrich_session.py on the same fixture, confirmed summary frontmatter added only to files missing it, and existing frontmatter left untouched.
• Lock‑file test: launched two instances of hourly_archive.py in parallel — second exited with “Another instance … is already running — exiting.” Same for archive_now.py vs hourly_archive.py.
• Scripts/vault_audit.py reports 0 broken links.
• Markdown fences in Dashboard.md and README.md remain balanced.

[1.17.0] — 2026-09-16 link

Added link

• 4 new Dashboard.md sections — the last 4 roadmap items, all shipped together:
  • “What’s changed since your last visit” — a delta banner comparing this note’s own frontmatter (written on every open) against current session/candidate counts, so you see “3 new sessions, 2 new candidates since you last opened this” instead of nothing.
  • “Storage & file counts” — live file count and total size via app.vault.getFiles(), plus a breakdown of the top 5 file types.
  • “Recently promoted” — parses Memory-Review/Consolidation-Log.md (written by consolidate_memory.py) and surfaces the last 5 decided candidates, so the review pipeline feels less like a black box.
  • “Top skills used” — parses Skills-Notes/Skill-to-Chat-Links.md (auto-generated by generate_skill_links.py) and ranks the top 5 skills by session count, working against both its tool-call mode and prose-mode output formats.
• All 4 sections degrade gracefully when their source file doesn’t exist yet (first-run/no-data messages, same pattern as every other Dashboard section).

Verified: all 11 dataviewjs blocks in Dashboard.md pass node --check; ran a full mock-harness pass with realistic fake data against every block (including a dedicated second-visit test for the delta banner, confirming it correctly detects 3 new sessions and 2 new candidates against saved frontmatter); Scripts/vault_audit.py — 0 broken links; markdown fences balanced (28 fence lines / 14 code blocks).

[1.16.0] — 2026-09-16 link

Fixed link

• Stale hardcoded skill count (roadmap item #4, last of the 4 known bugs) — Skills-Notes/README.md said “all 154 installed Hermes skills,” a number copied from one real vault’s install that goes wrong the moment any user installs or removes a skill. Reworded to “your installed Hermes skills” with no baked-in count, and pointed at hermes skills list for regenerating the index. Verified via Scripts/vault_audit.py: 0 broken links.

[1.15.0] — 2026-09-16 link

Fixed link

• Supermemory presented as required, not optional (roadmap item #3) — Scripts/consolidate_memory.py already handled missing Supermemory export files gracefully (its read() helper returns "" for a nonexistent path, so it never errored), but gave no signal to a non-Supermemory user that the pipeline was silently producing zero candidates every run. Added an explicit stdout message when neither Supermemory-All-Memory-Entries.md nor Supermemory-Explicit-Memories.md exists, pointing at the manual/Daily-sourced alternative documented in Memory-Review/TEMPLATE.md. Reworded the script’s docstring to describe Supermemory as one optional fact source, not the only one, and noted that automating non-Supermemory sources is a future roadmap item. All 6 existing Scripts/ci/test_consolidate_memory.py tests still pass; verified a clean 0-candidate run against a fresh vault with no Supermemory files (exit code 0, new message printed, valid Promotion-Candidates.md written).

[1.14.0] — 2026-09-16 link

Fixed link

• Hardcoded “4-agent team” assumption (roadmap item #2) — Welcome.md, Projects/README.md, Research/README.md, Skills-Notes/README.md, and Skills-Notes/Team-Profiles-Index.md all stated “the Hermes agent team (Codex, Ledger, Vox, Argus)” as a fact, contradicting Team-Profiles-Index.md’s own note that single-agent setups should delete it down to one row. Reworded all 5 files to say “your agent(s)” and relabeled the 4-name table in Team-Profiles-Index.md as an example/starter template rather than an assumed roster. Memory-Review/TEMPLATE.md similarly reworded to make agent attribution and Supermemory sourcing both explicitly optional. Verified via Scripts/vault_audit.py: 0 broken links.

[1.13.0] — 2026-09-16 link

Fixed link

• Missing “master hub note” (roadmap item #1) — Projects/Hermes-Agent-Vault-Setup.md was referenced by 6 files (Welcome.md, Projects/README.md, Research/README.md, Skills-Notes/README.md, SETUP.md, Memory-Review/TEMPLATE.md) as the vault’s master hub note, but never actually existed — every new user hit 6 dead links on first open. Wrote the actual file: covers folder structure, the memory pipeline, and status/decisions scaffolding, following the same frontmatter format as Templates/Project.md. Kept deliberately agent-count-agnostic (doesn’t hardcode a “4-agent team” assumption) so it doesn’t reintroduce roadmap item #2. Linked from MOC.md. Verified via Scripts/vault_audit.py: 0 broken links (down from all 6 references broken).

[1.12.0] — 2026-09-16 link

Added link

• User-Profile.md — creative pass. Added a TL;DR callout at the top, “talk to me like this / not like that” dialogue examples, ASCII trait sliders (directness/formality/risk tolerance), a new Availability & reaching you section (working hours, response-time expectations, notification channel), a new Decision authority section (unilateral action vs sign-off, optional/deletable for solo users), a scannable vibe-check table (mood → signal → what to do) replacing prose bullets, an “if I go quiet” playbook, a running jokes & call signs section, and a review cadence note in the Update log.
• README badge — new User Profile: documented badge linking to User-Profile.md, alongside the existing License/Obsidian/Hermes/OpenClaw/CI/Changelog badges.
• README “Known issues / Roadmap” section — checkbox list surfacing 4 known bugs (missing Projects/Hermes-Agent-Vault-Setup.md hub note referenced by 6 files, hardcoded 4-agent-team assumption in 5 files, Supermemory-only memory pipeline assumptions, a stale skill-count number) and 4 features under consideration (skill-usage breakdown, delta-since-last-visit banner, recent Memory-Review promotions log, storage/file-count stats) — tracked in the open instead of buried in commit history.

[1.11.0] — 2026-09-16 link

Added link

• Dashboard-Beta.md — opt-in, website-style card redesign of Dashboard.md. Same underlying data sources, same level of detail, nothing trimmed — just restyled as a card grid with a KPI strip up top instead of a plain vertical stack. Enable via Settings → Appearance → CSS snippets → toggle dashboard-beta. New sections beyond a straight restyle of the original:
  • Welcome header — live clock, time-of-day mood-based greeting, and an editable brand title/mantra (click-to-edit, localStorage-only — never written to any file, never committed).
  • Today’s focus — click-to-edit personal focus note, localStorage-only.
  • Quick actions — buttons that run real Obsidian commands (new Daily Review, open Memory-Review, open Vault Audit report, open Daily Timeline).
  • Session calendar — a month-grid with dots marking days that have archived Daily/ sessions; click a day to open it.
  • Weather — optional OpenWeatherMap widget. Requires pasting your own free API key into a settings popup on first use; the key lives only in that Obsidian window’s localStorage, never in a file, never in git.
  • On this day — surfaces Daily/ sessions from the same month/day in past years; renders nothing at all when there’s no match.
  • Ideas adapted from the Atlas and Komorebi dashboards in InlitX/Obsidian-Dashboard-Gallery (MIT) — kept the concepts, dropped what didn’t fit an agent-memory vault (e.g. Komorebi’s weather widget now needs an explicit city+key instead of a hardcoded default; personalization stays localStorage-only, never file-based).
  • Card-grid CSS ships as .obsidian/snippets/dashboard-beta.css, enabled by default in .obsidian/appearance.json.

[1.10.0] — 2026-09-14 link

Added link

• Activity heatmap on Dashboard.md — a GitHub-style, day-wise calendar heatmap (18 weeks × 7 days) showing token usage intensity per day, reusing the existing Skills-Notes/Token-Usage.log data (no new script or cron job needed). Color intensity scales toward the active theme’s accent color via 4 steps, so it matches whichever theme (Nemoclaw, Tokyo Night, etc.) is active. Hover any cell for the exact date, tokens, and session count. Built with plain colored <div> cells (not SVG) to avoid the Dataview parser fragility that broke the earlier token-trend graph in v1.5.1–v1.5.3.
• Scripts/seed_sample_data.py — opt-in sample/demo data generator (not shipped pre-populated, addresses the “empty vault on first open” problem from issue discussion around #9 without the staleness risk of committing fake data by default). Run it yourself to populate Daily/, Memory-Review/, Projects/, and Skills-Notes/Token-Usage.log with realistic-looking fake sessions/candidates/a project note, so a freshly installed vault’s Dashboard and graph view show a populated state instead of empty. Every generated file/entry is tagged sample: true (or [sample] in the token log) for easy removal later.

[1.9.1] — 2026-09-14 link

Changed link

• INSTALL_PROMPT.md — all 4 cron-job creation prompts now check before creating. Previously the wording only asked the agent to check for a conflicting archiving job before creating the hourly archiver; it didn’t explicitly tell it to skip creation if a job with the same name already existed for any of the 4 cron prompts (hourly archive, weekly vault audit, weekly agent performance, and the combined master install+cron prompt). Each now explicitly instructs: call cronjob_manage(action='list') first, and if a job with that name (or targeting the same vault Daily/ folder) already exists, don’t create a duplicate — just report that it’s already set up.

[1.9.0] — 2026-09-14 link

Added link

• Light-mode variant of the Nemoclaw theme (fixes #2) — added a full .theme-light block to .obsidian/themes/Nemoclaw/theme.css (all CSS variables + structural touches mirrored from the existing .theme-dark block, same NVIDIA-green accent darkened for AA contrast on white surfaces). Switch via Settings → Appearance → Base color scheme → Light while Nemoclaw is active. Theme manifest bumped to 1.1.0.
• Skills-Notes/Kanban-Usage.md (fixes #3) — new doc covering suggested default columns (Backlog/In Progress/Blocked/Done), how Kanban cards relate to Projects/*.md notes (link-don’t-duplicate pattern), and cross-platform behavior notes for the bundled obsidian-kanban plugin. Linked from MOC.md.

[1.8.0] — 2026-09-14 link

Added link

• OpenClaw badge in README, alongside the existing Hermes badge.

Fixed link

• update.sh / update.ps1 merge conflicts on User-Profile.md (fixes #5) — a user who fills in User-Profile.md before or independently of a template update could hit a hard “both-added” git merge conflict on every future update.sh run, since the template also ships its own (blank) copy of the same filename. Both update scripts now register a local-only merge=ours git attribute for User-Profile.md on every run: the first pull still delivers the blank scaffold to new vaults as normal, but every pull after that silently keeps the user’s local content on conflict instead of stopping with conflict markers. This is vault-local config (.git/info/attributes), never synced to the template repo itself.
• Conflict-resolution tip added to both update scripts’ failure output, and a new “Resolving update.sh/update.ps1 merge conflicts” section in CONTRIBUTING.md, covering the remaining conflict class (customized .obsidian/appearance.json / theme theme.css) that the merge driver doesn’t cover.

[1.7.1] — 2026-09-14 link

Fixed link

• Personal Data Guard workflow — the v1.7.0 release’s guard job failed on push because its allow-list was missing Daily/Timeline.md, Daily/Chat-Correlation.md, and Daily/.gitkeep, and its empty-tree fallback re-flagged every pre-existing template file as “new” whenever github.event.before was the null SHA. Completed the allow-list and changed the fallback to diff against the last commit instead of the empty tree.

[1.7.0] — 2026-09-14 link

Added link

• try.sh / try.ps1 — copy the template into a scratch temp directory and open it in Obsidian with zero commitment: no Hermes CLI registration, no cron setup, doesn’t touch your real Obsidian config. Delete the copy any time.
• uninstall.sh / uninstall.ps1 — cleanly remove a vault installation: unsets env.HERMES_VAULT_PATH if it points at the target vault, warns about any Hermes cron job that may still reference it, and optionally deletes the vault directory after a typed confirmation.
• CI: Personal Data Guard (.github/workflows/personal-data-guard.yml) — new workflow that fails a push/PR if it touches disallowed personal-content paths (Daily/, Memory-Review/, Research/, .smart-env/, plugin data.json files) outside the template’s own allowed exceptions (Memory-Review/TEMPLATE.md, Daily/README.md, Research/README.md). Automates what CONTRIBUTING.md previously only asked contributors to self-check.
• CI: consolidate_memory.py content-correctness tests (Scripts/ci/test_consolidate_memory.py) — 6 tests asserting on actual output content (not just “did a file get written”): exact-duplicate dedup, fuzzy near-duplicate dedup, secret-string scrubbing, false-positive check on normal facts, decided-candidate exclusion (regression test for the v1.1.0 hash-prefix bug), and state-file JSON validity.
• CI: extended runtime smoke-test — generate_skill_links.py now also runs against the fixture vault in the runtime-smoke-test job (alongside the existing vault_audit.py, agent_performance.py, hourly_archive.py coverage). pull_supermemory.py and trend_digest.py are intentionally excluded — the former needs a real/mocked network API + key, the latter needs sentence-transformers, a heavy ML dependency not worth adding to this lint job.
• CI badge in README, linking to the Lint Scripts Actions workflow.

Changed link

• CONTRIBUTING.md — notes that the Personal Data Guard CI job now automatically enforces the “never commit real vault content” rule, not just an honor-system reminder.

[1.6.0] — 2026-09-11 link

Added link

• User-Profile.md — new template note documenting the human user’s identity, communication preferences, technical environment, standing facts, current focus, interests, and boundaries for the agent. Includes a distinctive “What the agent has noticed about you” section, written by the agent (not the user) based on real interaction patterns — a mirror, not a form.
• Nemoclaw theme — new custom dark theme bundled in .obsidian/themes/Nemoclaw/, NVIDIA-inspired black background with signature green (#76b900) accents, monospace headings, and a subtle grid overlay. Now the template’s default theme (was Tokyo Night, still available as an alternative via Settings → Appearance).
• MOC.md and Dashboard.md link to User-Profile.md for discoverability.

[1.5.3] — 2026-09-11 link

Removed link

• Token usage trend graph — Removed the live SVG line graph visualization of token usage from Dashboard.md to simplify the token usage section.

[1.5.0] — 2026-09-11 link

Added link

• Token usage tracking — hourly_archive.py now extracts prompt/completion tokens from JSONL exports and maintains a running total in Skills-Notes/Token-Usage.log. Dashboard shows live token usage (today, all-time, 7-day trend, day-over-day).
• Token usage trend graph — Added live SVG line graph visualization of token usage over the last 30 days to Dashboard, showing daily totals with axis labels and data points.
• Vault integrity audit — vault_audit.py scans for broken wikilinks, stale Memory-Review entries (>30 days), duplicate sessions in manifest, and orphaned files. Writes Skills-Notes/Vault-Audit-Report.md. Dashboard shows summary.
• Agent performance dashboard — agent_performance.py analyzes session archives for skill usage frequency, tool call patterns, daily trends, top sessions by tokens. Writes Skills-Notes/Agent-Performance.md. Dashboard shows snapshot.
• Dashboard tiles for all three new features (Token usage, Vault audit, Agent performance) with click-through links.
• Skills-Notes/Token-Usage.log starter file.
• Updated INSTALL_PROMPT.md with optional weekly cron prompts for vault audit (Sunday 2 AM) and agent performance (Monday 3 AM).
• README script table updated with three new automation scripts.

Changed link

• hourly_archive.py — now dual-exports (markdown for vault + JSONL for token extraction), idempotent token logging.
• INSTALL_PROMPT.md — added optional weekly cron prompts for vault audit and agent performance.

[1.4.0] — 2026-09-10 link

Added link

• Dashboard.md — new “Vault health” section: flags a stale/dead hourly archiver (based on the newest Daily/ session’s timestamp) and a growing Memory-Review backlog, both read-only against files already in the vault.
• Scripts/hourly_archive.py — exports Hermes sessions from the last hour, reorganizes them into Daily/YYYY/MM/DD/, and maintains a deduped Daily/manifest.jsonl. Designed to be idempotent — safe to re-run.
• Scripts/pull_supermemory.py — pulls all memories for a Supermemory container tag and saves them as both a raw JSON dump and a human-readable markdown note.
• CHANGELOG.md and .github/ISSUE_TEMPLATE/ (bug report + feature request templates, Discussions link).
• INSTALL_PROMPT.md — new “Upgrade prompt” section for pulling template updates into an existing vault via a single Hermes paste (wraps update.sh/update.ps1).

Changed link

• .gitignore — added .smart-env/.

[1.3.0] — 2026-09-10 link

Added link

• In-vault update notifications: Dashboard.md now checks GitHub’s releases API live and shows a banner if a newer template version is available (fails soft if offline).
• VERSION file — tracks the installed template version; carried through fresh installs and future updates.
• update.sh / update.ps1 — safely pull template updates into an already-installed vault via a read-only git remote and 3-way merge, so hand-edited files surface as conflicts instead of being silently overwritten. Personal content (Daily/, Memory-Review/*, Projects/* beyond README, plugin data.json) stays protected via the vault’s existing .gitignore.

[1.2.0] — 2026-09-10 link

Added link

• Dashboard.md — single landing note with live Dataview views: active projects, open Memory-Review candidates, recent Daily sessions, installed-skills count.
• Projects/Projects.base — native Obsidian Bases table view over Projects/ (all projects grouped by status, active-only filter).
• Bundled Dataview plugin updated to 0.5.70.

Changed link

• README opening rewritten with a punchier hook instead of leading with a feature list; updated hero screenshot.
• Fixed dead links in MOC.md to Skills-Notes/Installed-Skills-Index and Team-Profiles-Index.

[1.1.0] — 2026-09-08 link

Added link

• Scripts/generate_skill_links.py — generates a Skill-to-Chat-Links index per session.
• Skills-Notes/Installed-Skills-Index.md and Team-Profiles-Index.md starter tables.

Fixed link

• Scripts/consolidate_memory.py — decided_hashes was comparing full 40-char hashes against 8-char prefix keys, so already-approved candidates kept reappearing on every run instead of staying excluded.

Changed link

• Consolidated New-TrendDigest.ps1 and Write-TrendDigest.ps1 (near-duplicate scripts) into one.

[1.0.0] — 2026-09-06 link

Added link

• Initial template release: Obsidian vault structure (Daily/, Memory-Review/, Projects/, Research/, Skills-Notes/), Kanban plugin, 3 note templates (Project, Daily-Review, Research-Note), Dataview query library, Canvases/Memory-Pipeline.canvas visual map.
• Cross-platform installers (install.sh / install.ps1), INSTALL_PROMPT.md for one-paste setup via a Hermes agent.
• CI (bash/python/powershell syntax checks), wiki, MIT license.