Repository navigation
Conversation
… workflow
- README linked ./SETUP.md twice; the file lives at docs/SETUP.md
- 'nine-phase' migration process actually has eleven phases (0-10)
- docs/AGENT-INSTRUCTIONS.md referenced .github/workflows/validate.yml three
times ('keep as-is, already correct') but the file did not exist. Added it:
frontmatter contract check, gitleaks secret scan, offline relative-link
check, advisory staleness report.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…typed learnings Implements the cross-validated two-tier memory pattern (OpenClaw workspace memory + Claude Code auto memory): a small always-loaded index, dated daily worklogs with today+yesterday loaded at session start, and an append-only learnings store typed pattern/pitfall/preference/architecture/tool (gstack /learn pattern). - memory/README.md is the full folder contract (loading conventions, entry shapes, prune rules, edge cases) - AGENTS.md: folder-map rows, Memory section, learnings-before-done rule - CLAUDE.md: memory added to session read order - Line cap for AGENTS.md raised 150 -> 200 across docs, matching Anthropic's published guidance for always-loaded files (AGENTS.md is now 194 lines) - validate-frontmatter.sh covers the curated memory files; daily worklogs stay exempt as time-series drafts Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ty, processes Management-system layer with generic, framework-agnostic artifact names and an explicit EOS/OKR/4DX mapping table (EOS terms attributed to EOS Worldwide; no affiliation claimed). Encodes YC doctrine: single north-star metric pinned atop a 5-15 measurable scorecard, definitions in the repo, values in live dashboards. - operating-system/README.md — folder contract + framework mapping + quality bar - vision.md — V/TO-compatible long-horizon layers; strategy.md stays quarterly - rocks.md — 3-7 quarterly goals, one owner, binary done; append-only quarter history - scorecard.md — definitions/owners/goals + dashboard links, never pasted values - accountability.md — seats with exactly one owner, cross-linked to people/ - processes/ — SOP contract (trigger/owner/steps/done/exceptions, <=500 words) - templates/process.md, templates/rock.md - weekly template + README gain an optional L10-compatible scorecard & rocks check - strategy.md <-> vision.md and ops.md <-> accountability.md cross-links - type enum extended (vision|rocks|scorecard|accountability|memory) in all four definition sites; routing table and migration tree updated Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- .claude/settings.json: PostToolUse(Write|Edit) hook runs the frontmatter validator; a violation feeds back to the agent immediately (Anthropic: prose is advisory, hooks are deterministic) - OPERATIONS.md: weekly scorecard & rocks check + CI-green gate; monthly learnings prune + memory archive; quarterly rocks rollover - ARCHITECTURE.md: principles 11 (handbook-first, GitLab) and 12 (advisory vs deterministic); reference list gains GitLab, Anthropic best practices, OpenClaw memory, gstack, Altman's Startup Playbook Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
agents-compat/openclaw/ maps the repo onto OpenClaw's bootstrap files: AGENTS.md and memory/ are used as-is (same two-tier convention), with SOUL.md / USER.md / TOOLS.md stubs to copy into the workspace root. Keeps OpenClaw's own safety rules: private repo, no secrets, third-party skills are untrusted code. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
README now lets a first-time human or agent absorb the whole system before descending: the promise statement, the three-layer diagram inline, a complete file map (one line per path), five reading paths by persona (new teammate / agent on a task / agent building from raw data / operator / OpenClaw user), the five core contracts, the operating cadence, alternatives considered (wiki / plain CLAUDE.md / vector memory / live-system mirror), an FAQ, and the EOS trademark attribution. INDEX.md gains the operator reading path, memory in the agent path, and domain rows for vision/rocks/scorecard/accountability/processes/learnings. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…mplate placeholders from link check
- gitleaks-action@v2 requires a paid license for organization accounts;
the pinned CLI binary is free and equivalent
- lychee flagged the {{URL}} scaffolding link in templates/weekly.md;
placeholder links are now excluded (raw + URL-encoded forms)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… complete folder docs Makes the template an instant starting package — clone, bootstrap, go: - Four first-party starter skills shipped as working SKILL.md (normalize-meeting, weekly-brief, customer-brief, update-wiki), symlinked into .claude/skills/ for Claude Code discovery; skills/README and migration docs updated accordingly - scripts/bootstrap.sh: interactive placeholder fill across the repo, scaffold date stamping, optional pre-commit hook install; SETUP.md step 2 now one command - docs/CLIENT-ONBOARDING.md: the consultancy engagement playbook — pre-kickoff checklist, 2h workshop agenda mapped to target files, ingestion-week rules (drafts only, client promotes to verified), handoff demo, retainer shape, engagement guardrails (client owns the truth; generic improvements flow back to the template) - Folder READMEs completed: templates/, docs/, scripts/, agents-compat/ — every folder now carries its contract - .github/PULL_REQUEST_TEMPLATE.md (the OPERATIONS good-PR checklist) and CODEOWNERS stub (founder owns all; finance/people/AGENTS.md/.github pinned) - docs/TEMPLATE-CHANGELOG.md: template versioning (v1.1) so client repos can diff and pull future template improvements - README: file map, quick start, FAQ updated Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Member
Author
|
Second wave pushed — turnkey package (template v1.1):
After merge, suggest tagging 🤖 Generated with Claude Code |
lychee resolves relative links inside .claude/skills/* against the symlink path instead of the real skills/ location. The canonical copies under skills/ remain fully checked. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…to be reshaped meetings/ now routes notes into <category>/YYYY-MM-DD-<topic>.md across a YC/EOS-shaped default cadence: standups, leadership (L10-compatible), one-on-ones, customers (the talk-to-users record), product, marketing, board, planning (quarterly pulse). Each category README states what belongs, its distill focus, propagation rules to canonical files, and a hard boundary where one applies (no comp/perf content in 1:1s, no term details in board notes). The taxonomy is explicitly a starting point: migration Phase 8 now begins by fitting it to the company's actual calendar — delete unused categories, rename/add what the real cadence demands. Folders reflect real meetings, never aspiration. Cross-references updated: AGENTS.md folder map, README file map, routing table, migration tree + Phase 8, naming examples, weekly/README globs + example, normalize-meeting and weekly-brief skills, INDEX, template changelog. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The interaction model is: humans talk to the agent (Slack agent, Claude Code); the agent reads and writes the repo; humans review PRs on GitHub. Nobody is expected to browse folders, run a docs site, or edit Markdown in an editor. - README audience bullets + diagram + reading path rewritten agent-first; 'the agent is the interface; the repo is its memory' - ARCHITECTURE diagram + 'Renderable' rationale replaced (agent-readable everywhere; no docs site to maintain) - All MkDocs/Obsidian/docs-site/editor mentions removed (SETUP prereqs, OPERATIONS weekly ritual -> agent spot-check question, AGENT-INSTRUCTIONS do-not-create list + handoff + next steps, people/README, .gitignore) - Notion's role clarified: stays as the working-docs surface (live system via MCP); durable truth lives here Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…-native skill creation, language rule Founder feedback round (template v1.2): - operating-system/ -> management/; rocks.md -> goals.md; scorecard.md -> kpis.md. Plain-English names primary; EOS terms (Rocks, Scorecard, L10) kept as equivalents in the mapping table. Types rocks|scorecard -> goals|kpis across all enum sites and frontmatter. - docs/ -> guides/. - Meeting taxonomy 8 -> 6: board/ and planning/ removed from the default (not usual for most companies); both listed as add-when-relevant categories with the planning propagation duty preserved. - Agents codify repetition natively: AGENTS.md rule (3rd occurrence of a workflow -> propose a skill via PR), 'codify repetition yourself' section in skills/README with detection signals, monthly skill-candidate sweep in OPERATIONS. - templates/README states what templates are (file skeletons for the repo's own types) and are not (business-document templates — that know-how lives in the generation guides inside the files). - people/ scope includes freelancers/contractors (relationship: freelancer). - Language rule: English is the operating language (compiled truth, summaries, frontmatter); verbatim quotes, transcript excerpts, and local-market brand voice examples stay in their original language. Stated in playbook, AGENTS.md, AGENT-INSTRUCTIONS, brand.md guide, meetings contract, README FAQ. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…udit Audit: cold-start agent simulation, cold-start founder simulation, cross-doc consistency forensics, SME gap analysis vs a production SSOT, and executable verification of every script/hook/workflow. ~45 findings, all addressed: Contract fixes - Weekly briefs removed from auto-commit lists (AGENTS.md + AGENT-INSTRUCTIONS) — they are PR + human review only; was the worst contradiction (canonical contract authorized pushing the one document whose folder contract forbids auto-merge) - ops | finance added to the canonical type enum (two shipped root files violated AGENTS.md's own contract) - Wikilink semantics defined (root-relative, no .md, agent convention, not validated by CI); broken worked examples fixed (meetings category segment); ../-relative wikilinks normalized - Precedence rule (folder README > AGENTS.md > guides) + [VERIFY] / [INFERENCE] uncertainty tags with a weekly resolution ritual New files - GLOSSARY.md (every system term incl. ADR, IDS, L10, MCP, wikilink) - guides/HUMAN-JOB.md (the founder's whole role, click-by-click PR review) - guides/FAQ-AND-FAILURE-MODES.md (judgment calls, CI failures, anti-patterns) - LICENSE (MIT — README claimed it; file didn't exist), SECURITY.md (burned-credential rotate-then-use), CONTRIBUTING.md - Makefile (make verify = CI parity) + scripts/check-links.sh - sales/pipeline.md + sales/pricing.md scaffolds (INDEX/README routed to files that didn't exist) Script bugs (verified by re-running end-to-end) - bootstrap.sh: re-runs aborted silently under set -euo pipefail when a placeholder was already filled; date-stamp regex missed 2026-06-12 scaffold dates; leftover count flagged intentional template scaffolding - pre-commit dirs aligned with validate-frontmatter.sh; hook command anchored to $CLAUDE_PROJECT_DIR Docs reconciled - AGENT-INSTRUCTIONS: eight→six categories, board/ example replaced, template-ships-the-structure framing, complete target tree, archive rule scoped to migration time, make verify in Phase 10 + checklist - SETUP: agent-driven path first, gh prereq, seed list reconciled (finance.md, management/, memory, inbox), /skill x → /x unified, honest branch-protection note for free private repos - Routing rows: glossary, legal pointers (new ops.md section), job candidates → ATS, content calendar → live systems; github added to source_of_truth enums; strategy↔goals boundary stated; people alumni → status: archived; templates gained REQUIRED fields Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ommits The mechanics existed (git + PR-gated truth, semantic commits, append-only timelines, last_verified) but two things were missing per founder QA: nothing required an agent to identify itself in commits, and the versioning story wasn't discoverable. - AGENTS.md 'Sign your work' rule: semantic commit titles, the why in the body, and a Co-Authored-By: <agent> trailer on every agent-authored commit — never impersonate the human. (Made room by dropping the operating-rules duplicate of 'one idea per file', which Naming conventions already states; file stays at the 200-line cap.) - README: new 'Versioning & audit trail — built in, not bolted on' section (what/why/who-human-or-agent/when, History/Blame, the no-hand-maintained-changelog doctrine and its one exception) - HUMAN-JOB: 'Seeing the history' for non-technical founders - GLOSSARY: audit trail + blame entries; FAQ: 'where's the changelog?' - OPERATIONS good-PR list + PR template + CONTRIBUTING: provenance line Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Closes the last acknowledged gap: the README diagram promised 'ask in Slack' with no wiring path, and nothing told a team how non-technical people read/edit the SSOT. - guides/TEAM-ACCESS.md: three tiers (1 ask&edit via the Slack agent — everyone, no GitHub account needed; 2 browse read-only published Notion views; 3 review&merge on GitHub — 1–2 maintainers). The one-way-out / never-two-way rule with the dual-write/bypassed-gate/corruption reasoning. Published-view stamped-header convention driven off the audience: frontmatter field. Slack-agent wiring options (Claude in Slack / OpenClaw-in-a-channel / n8n webhook bridge) with read-broad, write-as-PR invariants. - Wired into README (teammate reading path + FAQ + file map), guides/README, ARCHITECTURE (deliberately-NOT now names two-way sync), HUMAN-JOB. Changelog v1.4. Mental model: Notion is the noticeboard, Slack is the reception desk, GitHub is the vault. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Linear was handled correctly throughout the architecture but absent from the team-facing access doc. Added a 'Where Linear fits' section: the durable-vs-live layering, the 'is this still true next week?' test, the repo↔Linear handoffs (decisions spawn tasks, product.md links the roadmap, shipped work flows back, the weekly brief joins both), and the rule that neither mirrors the other. Extended the mental model (Linear = the job board) and the quick decision guide. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Slack is the room; the agent living in it is who you actually talk to — it reads the vault, drafts changes, opens PRs. Metaphor and Tier-1 row updated; kept runtime-agnostic (Claude in Slack / OpenClaw bot / n8n bridge) with a pointer to the wiring section. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Structure-conventions review: the 'no numbered prefixes, clean names' rule existed in AGENT-INSTRUCTIONS but its reasoning was undocumented — which made the convention unabsorbable and the divergence from numbered-folder SSOTs (e.g. 00_/01_ schemes) look arbitrary. ARCHITECTURE.md now has a 'Folder & file naming' section: agents route by name + frontmatter type: not tree position; guessable names are self-locating (biggest absorbability win); knowledge bases have no inherent sequence (INDEX/reading-paths supply order where it matters); clean names absorb new folders without renumber cascades. Plus the plural-folder/singular-root-file + kebab/ISO conventions, and an honest note on when numbered prefixes (Johnny.Decimal) ARE right (human-browsed trees, curricula) — a different trade-off, not a contradiction. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…(v1.6) A seventh root knowledge file for the durable administrative facts a company needs day to day — the things you paste into an invoice or a contract — which previously had no clean home (company.md is the pitch, finance.md is money-truth pointers, ops.md only held a doc-location stub). legal.md holds: legal entity (registered name, legal form, registration no., tax/VAT IDs, incorporation date, jurisdiction), registered + invoicing + trading addresses, company structure (entities, high-level ownership, directors/signatories), key registrations (EORI/DUNS/licences), receiving bank/IBAN if shareable, and pointers to the underlying legal documents. Mostly low-sensitivity public-register data, so it lives in the repo; cap table / exact share % / online-banking creds explicitly stay out (finance.md → encrypted store). Wired through everything: type: legal in all three enum sites, routing rows (entity/VAT/invoicing-address/structure → legal.md; legal-documents row retargeted from ops.md to legal.md), README file map, INDEX domain table, SETUP seed list, validate-frontmatter ROOT_FILES, migration tree + Phase-4 order + six→seven-root counts, full playbook fill-guide. The old ops.md 'Legal pointers' stub consolidated into a cross-link (no dup). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ment 3-agent diligence (steelman for/against numbered folders + empirical survey of real SSOT repos) confirmed clean taxonomy names are best practice for an agent-first company SSOT. Folded the findings into ARCHITECTURE.md § Folder & file naming: - documents-vs-taxonomy distinction (we DO date/number sequential documents — ADRs, decisions, weekly; we DON'T number taxonomy folders — the case against numbering is scoped to taxonomy prefixes) - field convention as evidence (5/5 public company handbooks + ~90% of agent-context repos use clean names; numbering's home turf is human-browsed PKM vaults / rendered-docs sidebar sort keys / individual decision docs — none transfer to an agent-first git SSOT) - honest framing: numbered taxonomies are a marginal cost, not a failure (perfectly agent-legible — a human-browsed repo may reasonably number; different audience, not a contradiction) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Member
Author
|
Cherry-picked what main needed from this branch on 2026-09-09 (LICENSE, validate.yml, CODEOWNERS, PR template, CONTRIBUTING, SECURITY, re-pointed to docs/). The branch's guides/ layout, Makefile and memory/ tree diverge from main; closing rather than merging. Reopen a targeted PR for any remaining piece. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this is
Full implementation of the research-backed upgrade blueprint (deep-research pass: 22 sources, 24/25 claims verified 3-0 against primary sources — GitLab handbook-first, OpenClaw workspace & memory docs, Anthropic Claude Code best practices, gstack, Altman's Startup Playbook).
Changes by phase
P0 — defects
./SETUP.md→./docs/SETUP.md(README linked a non-existent path twice).github/workflows/validate.ymlthat three docs already claimed existed (frontmatter contract, gitleaks, offline link check, advisory staleness report)P1 — agent memory layer (
memory/)MEMORY.mdindex (hard cap 200 lines / 25 KB, always loaded) +YYYY-MM-DD.mddaily worklogs (today+yesterday at session start) +learnings.mdtyped lessons (pattern/pitfall/preference/architecture/tool — the gstack/learnshape)P2 — operating-system pack (
operating-system/)vision.md(V/TO-compatible long horizons),rocks.md(3–7 quarterly goals, one owner, binary done, append-only quarter history),scorecard.md(5–15 weekly measurable definitions + dashboard links — never values; north-star pinned on top per YC doctrine),accountability.md(one name per seat),processes/(SOP contract)process.md,rock.md)P3 — deterministic enforcement + rituals
.claude/settings.json: PostToolUse(Write|Edit) hook runs the frontmatter validator (advisory prose → deterministic hook, per Anthropic)P4 — README as the exhaustive front door
P5 — OpenClaw compatibility (
agents-compat/openclaw/)memory/work unmodified as an OpenClaw workspaceVerification
scripts/validate-frontmatter.sh✅ on every commitCaveat for reviewer
EOS artifact internals (V/TO sections, L10 agenda order) follow conventional knowledge of the framework — the deep-research round verified everything except primary EOS sources. Before promoting the template publicly, sanity-check
operating-system/wording against Wickman's Traction; trademark attribution is already in place (README + operating-system/README).🤖 Generated with Claude Code