v1.6.1 — README/landing parity, honest claims, hermetic test suites

Fixed

• README: the directory tree listed files that do not exist
  (docs/publicacion-github.md, docs/difusion/); skill count (15), suite
  count (7) and assert count (178) now match the tree; ES script aliases
  documented; test badge derived from the real suites.
• Landing (docs/index.html): language note aligned with the README (factory
  in English, working language chosen at init); claude step added between
  init-system.sh and /repo-add.
• Comparison table: replaced the vague "limited" cells with what each tool
  actually offers for brownfield; tagline softened from "cannot lie" to
  "derived from real sources, with provenance seals" (README, landing, banner).
• test-dora.sh: fixture commits carry an explicit git identity, so the suite
  passes on clean runners without user.name/user.email.
• test-docs.sh: the smoke run over the real workspace now writes to a temp
  copy; running the suite no longer modifies tracked docs/*.html.
• generate-as-is.sh: structure step no longer aborts on empty repos under
  pipefail; vendored contracts/ are excluded from consumption detection.
• .gitignore: harness/.venv/.

v1.6.0 — /context-report pre-empts committee-recurring findings

Changed — /context-report skill & documentation standard

Lessons distilled from generating several application context reports, encoded so future runs clear the review committee on the first pass:

• Valid CML — declare every context used in a relation (incl. external systems), keep the upstream/downstream direction right, one downstream strategy per upstream (ACL or Conformist, not both).
• ADRs as decision atoms — Context, Options considered with trade-offs, Decision, Consequences; outcome-only prose fails.
• Given/When/Then verifiable quality scenarios with anchor + RED/AMBER/green verdict.
• Cross-document count reconciliation — two authoritative docs must never contradict each other on the same seal; reconcile or fix.
• Complete Annex A inventories — promise N, list N.
• Honest severity calibration — a parametrized query with a constant-bounded table name is a mitigated smell, not a critical injection; reward calibration, reject both alarmism and omission.

Documentation standard (§5 ADR) gains a Common pitfalls note naming the three most frequent failures.

Fully generic — no example-stack or client/domain references. All 7 self-test suites green (178 asserts).

v1.5.0 — repository file structure as derived documentation

Added

• scripts/repo-tree.sh <path> [depth] — deterministic file-structure tree: tracked files only (respects .gitignore), recursive per-directory counts, root files listed, find fallback for non-git dirs. Suite test-repo-tree.sh (17 asserts, committed red before the implementation).
• System architecture document: new derived section 5.2.1 Repository file structure — generate-architecture.sh injects a depth-2 tree per registered repo on every regeneration ({{FILE_TREES}} placeholder in the template).
• Documentation standard: new required Annex E — repository file structure (derived tree + curated directory → role → section annotations, development-view complement).
• /context-report skill: Annex E format encoded (root at depth 2, source dir at first level only, annotation table), structure-driven finding mining (loose migration scripts, deprecated infra trees, tracked tool residues), the E.4 sibling-repos pattern for standalone reports, and the system-doc regeneration step. Framework-neutral wording throughout.

All 7 self-test suites green (178 asserts).

v1.4.0 — /context-report + remote fleet support

Added — application context reports (/context-report)

• New skill /context-report <repo> [--update]: generates the maximum-detail application architecture & context report (arc42 + C4 + 42010) — parallel file:line mining, graph queries with count reconciliation, obsolescence research with sources, a business-flow traceability matrix (UI route → front module → service → consumed domains → upstream route), an end-to-end view with a per-hop table, sequence diagrams, enriched ADRs, numbered verifiable quality scenarios, and a mandatory adversarial review committee (architecture + quality agents) before the human gate.
• Corporate HTML twin for application reports: templates/knowledge-architecture-app.html (light/dark palette, semantic category colors, provenance-box header) + scripts/app-architecture-html.py (reuses the md→html converter embedded in generate-architecture.sh). New suite test-app-architecture-html.sh (14 asserts, committed red before the implementation).

Added — remote fleet support (codebase-memory-pg)

• docs/codebase-memory-setup.md: new Remote fleet — Cloud Run deployment section — client configuration for a cloud-deployed fleet facade, including the X-Flota-Authorization header contract (Cloud Run strips the Authorization token signature after its IAM check), audience and identity (sub) rules, the project-naming requirement, and the audited remote seeding flow (/admin/apply → staging upload → /publish).

All 6 self-test suites green (161 asserts).

v1.3.0

v1.3.0 — AI-consumable architecture documentation standard

Adds the standard the factory follows to produce maximum-detail, AI-consumable application architecture documentation, synthesized from the major frameworks and consultancies and adversarially fact-checked (two deep-research passes, 25+ claims verified 3-0).

Added

• docs/architecture-documentation-standard.md — layers arc42 + C4 + ISO/IEC/IEEE 42010 with TOGAF (traceability matrices, ADD/ARS), Zachman (W5H coverage), Rozanski & Woods (Information/Operational viewpoints + quality perspectives), DDD / Context Mapper (CML), 4+1 (Kruchten), Google Design Docs & SRE, AWS/Azure Well-Architected, and Gartner Pace-Layering — each cited. Includes a unified mapping table (framework element → arc42 section → what to deepen for AI) and the enriched ADR format.
• The knowledge-architecture templates realize the standard:
  • Front-matter metadata: standard, viewpoints, quality_attributes, zachman_coverage.
  • New Information view (data), Integration & APIs view, TOGAF traceability matrices, Operational view (SRE).
  • Enriched §9 ADRs (Nygard + Google: goals/non-goals, alternatives considered, confidence; append-only, immutable).
  • Annex D · traceability matrix (requirement ↔ ADR ↔ building block ↔ file:line ↔ assertion).
  • The generator needs no change — the new curated sections are narrative-driven.

Changed

• CLAUDE.md, README and the <prefix>-architecture skill point at the standard; test-architecture.sh covers the new sections (147 asserts total).

Coverage note

ArchiMate and BCG/Accenture/McKinsey thought leadership were targeted but not verifiable (primary sources behind login/paywall); documented as gaps, not omissions.

Full changelog: https://github.com/mmaripanguec/dev-sdd-kit/blob/main/CHANGELOG.md

v1.2.0

v1.2.0 — Automatic code-graph indexing for the fleet/Postgres facade

Turns the codebase-memory handling from guidance into real, configurable
code. Previously, onboarding a repo against a read-only fleet/Postgres facade
surfaced -32601 (index_repository has no equivalent in the fleet graph) and
left indexing as a manual, documented step. Now the factory seeds the repo and
wires the workspace MCP config automatically when the fleet is configured —
while still never blocking onboarding.

Why

The fleet/Postgres facade is read-only and does not expose the MCP write tool
index_repository (returning -32601 is expected, not a bug). The direct
engine indexes via the /repo-add skill; the facade needs an out-of-band seed.
v1.2.0 automates that seed + MCP wiring so both modes "just work".

Added

• scripts/codebase-memory.sh — the indexing driver (bash 3.2, no deps):
  • index <repo> — seed the repo into the fleet via its CLI, then upsert a
    cbm-<repo> HTTP entry into a gitignored workspace .mcp.json.
  • mcp-config [repo…] — (re)write .mcp.json entries for the given repos
    (or all registered repos).
  • mode — print the resolved mode (fleet / direct / off).
• repo-add.sh integration — after registering a repo, it calls
  codebase-memory.sh index <name> (non-blocking). With the fleet configured,
  onboarding indexes automatically; otherwise it prints direct-engine guidance.
• scripts/tests/test-codebase-memory.sh — 5th self-test suite. Total is
  now 141 asserts across 5 suites, all green.

Configuration (.env, see .env.example)

Variable	Purpose
CBM_MODE	auto (default) · fleet · direct · off
CBM_FLEET_SEED	Seed command template; {repo} → repos/<name>. Its presence turns auto into fleet.
CBM_FLEET_URL	Fleet gateway base URL (e.g. http://127.0.0.1:8787)
CBM_FLEET_TOKEN	Bearer token for the fleet gateway
CBM_FLEET_PROJECT	Project-id template; {name} / {abspath_dashes} (default: {abspath_dashes})

Nothing is hardcoded — every fleet specific lives in .env.

Usage

# One-time: configure the fleet in .env, then onboarding indexes automatically:
./scripts/repo-add.sh https://github.com/org/repo.git   # seeds + wires .mcp.json

# Or drive it directly:
./scripts/codebase-memory.sh index <repo>    # seed one repo + wire .mcp.json
./scripts/codebase-memory.sh mcp-config       # (re)write .mcp.json for all repos
./scripts/codebase-memory.sh mode             # show the resolved mode

Behavior

• Non-blocking: if the seed command fails or the fleet is down, it warns and
  continues; the as-is map, context packs and architecture document are still
  built from static analysis ("indexing pending").
• Read-your-writes: re-seed after structural code changes; the graph does
  not refresh itself.

Changed

• docs/codebase-memory-setup.md documents the automatic path and the .env
  configuration (direct engine vs fleet facade).
• .gitignore now ignores .mcp.json (machine-specific paths + token).
• .env.example gains the CBM_* block; the /repo-add skill notes the seed +
  .mcp.json wiring happen automatically when configured; README badge 119→141.

Upgrade notes

• No breaking changes. If you were wiring .mcp.json by hand, you can now set
  CBM_* in .env and let /repo-add do it. Ensure .mcp.json is gitignored
  (it now is by default).

Full changelog: https://github.com/mmaripanguec/dev-sdd-kit/blob/main/CHANGELOG.md

v1.1.2

Cleanup release (docs only, no code changes).

Removed

• The last example-system reference from the public tree; the operating guide now describes repos.yaml generically (created at instantiation via init-system.sh, populated with /repo-add), which also fixes a stale note.

Full changelog: https://github.com/mmaripanguec/dev-sdd-kit/blob/main/CHANGELOG.md

v1.1.1

Documentation follow-up for the v1.1.0 architecture-as-context capability (no code changes).

Changed

• Operating guide (ES) — documents the generated arc42+C4 architecture document as agent context, the docs-first policy and the <prefix>-architecture skill, the codebase-memory fleet/Postgres facade troubleshooting (-32601), and structure/shared-memory/governance updates.
• README — front-page highlight of the AS-IS architecture as agent context, plus updated comparison-table rows (brownfield grounding, derived docs).

Full changelog: https://github.com/mmaripanguec/dev-sdd-kit/blob/main/CHANGELOG.md

v1.1.0

Knowledge-as-context architecture documents and clearer code-graph setup.

Added

• AS-IS architecture document as knowledge (arc42 + C4): generic templates and a generator (scripts/generate-architecture.sh) that writes knowledge/architecture/<system>.md and a self-contained .html twin (Mermaid), combining a curated per-system narrative with data derived from the code (topology, dependencies, counts, seals).
• The generator runs after every /repo-add indexing (hook in generate-as-is.sh).
• /system-map authors the narrative, generates the document, and creates a <prefix>-architecture context skill encoding a docs-first policy (answer from the document; read code only for undocumented gaps).
• docs/codebase-memory-setup.md; self-test suite (scripts/tests/test-architecture.sh).

Changed

• /repo-add documents both codebase-memory modes and treats the fleet/Postgres-facade -32601 (index_repository unavailable) as expected, not a hard failure.
• CLAUDE.md, .env.example, init-system.sh, README and CHANGELOG updated.

Validation

• 4 self-test suites green (119 asserts, up from 99).

Full changelog: https://github.com/mmaripanguec/dev-sdd-kit/blob/main/CHANGELOG.md

v1.0.0 — first public release

First public release of the spec-driven factory template. Highlights: full F0–F9 lifecycle with six permission-backed human gates; structured clarification, cross-artifact consistency and code↔spec convergence; derived as-is map, DORA metrics and bilingual architecture docs (sealed, CI-checked); a real, recorded demo (ADK + Gemini customer assistant) with human-in-the-loop approval; 99 TDD-first asserts across three self-test suites. Commands are English; each workspace picks its working language (en/es) at init. See CHANGELOG.md for the full list.