feat: genericize into self-hostable OSS project

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.

[datj9]

Summary

Turns tech-bytes into a real OSS project: anyone can fork, edit one config file, plug in their LLM key (OpenAI / Anthropic / local), and launch with zero AWS. GitHub Actions builds the digest on a cron and publishes to GitHub Pages. AWS/CDK stays as an optional advanced path.

Plan: ~/.claude/plans/purring-kindling-cray.md

What changed

Pipeline (lambdas/ → pipeline/)

• Pluggable LLM (pipeline/providers/): openai, anthropic, local (OpenAI-compatible: Ollama, LM Studio, Groq, OpenRouter) behind one Protocol. Selected by llm.provider in config; keys from env, SSM fallback on AWS.
• Pluggable storage (pipeline/storage/): local_fs (default, writes data/) and s3 (AWS path) behind one Protocol. boto3 fully lazy — free path imports no AWS libs.
• Single CLI: python -m pipeline.run [all|release_radar|hn_digest|gh_trending|email_digest].
• Single config: config/techbytes.config.yml owns sources + branding + LLM choice. Icons are emoji literals in config (the site reads them directly; the old hardcoded categoryIcons map is gone).
• CloudWatch metrics no-op outside AWS. trigger_rebuild no-op on the free path.

Site (Astro)

• Branding read from config at build time; PUBLIC_* env overrides for CI.
• Pages base path auto-detected from GITHUB_REPOSITORY for project Pages sites.
• Data dir resolved from $DATA_DIR (default ../data).
• contribute.astro repo URL resolved from env — no hardcoded datj9/tech-bytes.

Workflows

• digest.yml (NEW) — free path: cron → run pipeline → commit data/ → build → deploy Pages. No AWS.
• deploy-aws.yml (renamed from deploy.yml) — optional AWS path; reads domain/zone/repo from vars/context.
• preview.yml — ruff + pytest + site build + best-effort cdk synth on every PR.

Infra (CDK, optional)

• Fully parameterized: domain.name, hosted.zone.name, github.repo, ssm.prefix from context. No hardcoded values; clear error if missing. See infra/README.md.

OSS scaffolding

• LICENSE (MIT), README.md (60-second quickstart + config reference + provider setup), CONTRIBUTING.md, regenerated .env.example.
• All personal identifiers stripped (domain, AWS profile, repo slug, SSM paths, rebuild URL). Grep gate clean across source and build output.

Decisions (locked at execution)

• License: MIT
• Brand: keeps "Tech Bytes" as the demo default (forkers override via config)
• Dir rename: lambdas/ → pipeline/ (git mv preserves history)
• Anthropic default model: claude-sonnet-4-6

Verification

All green locally:

Check	Result
ruff check pipeline/	clean
pytest pipeline/tests/	34/34 pass
pnpm run build (site)	success
tsc --noEmit (infra)	clean
python -m pipeline.run release_radar (smoke, local storage, fake key)	writes data/release-radar.json, no boto3 import
Grep gate finaldivision|datj9|super-personal in site/dist/	clean
Grep gate in source (excluding intentional default /tech-bytes/ SSM namespace)	clean

preview.yml runs ruff + pytest + site build + cdk synth on this PR.

Test plan for reviewer

1. Free path (local):cp config/techbytes.config.example.yml config/techbytes.config.yml → set OPENAI_API_KEY (or LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY) → python -m pipeline.run all → confirm data/*.json written, no AWS deps required.
2. Site:cd site && npm i && npm run build → confirm generic branding, no finaldivision / hardcoded icons in dist/.
3. Provider swap: toggle llm.provider between openai / anthropic / local in config; confirm the factory picks the right provider and default model.
4. Fork simulation: set GITHUB_REPOSITORY=myuser/myrepo before pnpm run build; confirm Pages base path and contribute page use the fork's identity.

Breaking changes

• config/technologies.yml removed → migrate to config/techbytes.config.yml (see example).
• data/release-radar.json icon field now an emoji literal (was a categoryIcons key string). Seed data migrated; first pipeline run regenerates everything.
• Lambda handler paths changed from release_radar.handler.handler to pipeline.release_radar.handler.handler — CDK bundling updated to match. Existing AWS deployments need cdk deploy to pick up the new handler paths.