Turn Codex, Claude Code and Git history into evidence-backed developer recaps, standups and work reports.
1
stars
78
commits
TypeScript
primary language
Sep 14, 2026
updated
Your coding history already knows what you did. DevRecap turns it into a recap.
Codex + Claude + Git → evidence-backed developer recaps.
Now available in the OpenAI Plugins Directory.
Install with codex plugin add devrecap@openai-curated, then run $devrecap.
You know the question.
“What did I actually work on this week?”
Then you open GitHub, scroll through commits, search old terminal sessions, try to remember what was finished, what was only investigated, and what is still in progress.
DevRecap reconstructs that work for you.
It reads only the local coding history and Git evidence you explicitly authorize, correlates the evidence, separates completed work from unfinished work, and lets the host AI turn those facts into a polished work report.
Not a timesheet. Not background monitoring. A recap you explicitly ask for when you need it.
DevRecap 0.3.0 is publicly available through the OpenAI Plugins Directory.
Install the approved public release directly from the OpenAI curated marketplace:
codex plugin add devrecap@openai-curated
Start or restart Codex, then invoke the Skill:
$devrecap
You can also install it interactively:
/plugins
Search for DevRecap, install it, start a new thread, then run:
$devrecap
If you want the latest version from this repository instead of the curated marketplace snapshot:
codex plugin marketplace add jquinteiroo/devrecap --ref main
codex plugin add devrecap@devrecap-marketplace
For most users, devrecap@openai-curated is the recommended install because it uses the reviewed public release.
Then ask naturally:
Help me remember what I worked on this week.
Or:
Me ajuda a lembrar tudo que trabalhei nas últimas duas semanas.
Quero um relatório detalhado em português.
On first use, DevRecap requires explicit permission before reading local Codex history, Claude history, or Git evidence.
DevRecap is designed to produce a recap you can actually use in a daily, weekly review, 1:1, sprint review, handoff, or performance conversation.
A typical report groups work into meaningful fronts such as:
Executive summary
Main focus
• Contract generation and document workflow
• Tournament API consistency and filtering
Confirmed deliveries
• Evidence-backed completed work
Investigations
• Work that was explored but not proven complete
In progress
• Active work without completion evidence
The final HTML report is editorial rather than commit-centric: context and narrative first, raw evidence available when you need to inspect it.
Evidence first. If DevRecap cannot support a claim from the collected evidence, it must not present it as completed.
A file edit is not automatically a finished feature.
A command is not automatically a delivered fix.
A conversation is not automatically proof that something shipped.
DevRecap combines session evidence with Git history and keeps uncertain work uncertain.
Without DevRecap:
I think I worked on the PDF generation flow...
I also fixed something in the product-selection flow.
And I investigated a bit the issues with the reports.
With DevRecap:
This week
Completed
• Improved the contract-generation flow and correlated the work with Git commits.
• Fixed frontend behavior in the product-selection journey.
Investigated
• Report-generation issues that still lacked completion evidence.
In progress
• New user creation in the admin panel.
The goal is not to make your week sound busier.
The goal is to make it accurate, useful, and easy to present.
You invoke $devrecap
↓
DevRecap resolves the requested period
↓
Authorized Codex + Claude sessions
+
Read-only Git evidence
↓
Normalize events
↓
Build meaningful Activities and Workstreams
↓
Correlate session activity with Git evidence
↓
Build an evidence-only analysis contract
↓
Codex / Claude writes the narrative
↓
DevRecap validates IDs, status and completion claims
↓
Editorial HTML report
The coding agent is the writing layer.
The DevRecap CLI is the factual layer.
That separation is intentional.
DevRecap is explicit-invocation only.
It does not automatically scan your repositories, watch your filesystem, or monitor your work in the background.
Before local collection, the user chooses which sources are allowed:
Codex history optional
Claude history optional
Git optional, read-only
Disabled sources are not invoked.
Git collection is guarded as read-only. DevRecap does not commit, push, checkout, reset, merge, rebase, or modify repositories as part of report generation.
See PRIVACY.md for the full privacy model.
The public Codex plugin ships the canonical Skill at:
skills/devrecap/SKILL.md
Claude support is also available in the repository:
.claude/skills/devrecap/SKILL.md
The AI-first flow is:
prepare
↓
.devrecap/run.json
↓
host AI analyzes contract.facts only
↓
.devrecap/analysis.json
↓
validate
↓
render
↓
reports/derecap-YYYY-MM-DD.html
The agent is instructed to:
in_progress, blocked, or unknown work to completed;DevRecap can collect authorized local Codex session history inside the requested time range.
Authorized Claude coding sessions can be included in the same recap, allowing work performed across agents to appear in one report.
DevRecap correlates relevant work with read-only Git evidence.
Commit evidence is especially useful for distinguishing:
worked on it
from:
there is evidence that this work was completed
The public directory is the easiest way to use DevRecap. Clone the repository only if you want to develop or inspect the project locally.
Clone:
git clone https://github.com/jquinteiroo/devrecap.git
cd devrecap
npm run setup
Run the CLI from the repository:
npm run recap -- "this week"
Useful commands:
npm run setup
npm test
npm run typecheck
npm run smoke
npm run dev
The direct CLI can still generate a deterministic recap without host-AI synthesis:
devrecap today
devrecap week
devrecap remember "last 14 days"
The preferred experience for polished reports is the Skill, because Codex or Claude can turn the structured Activities into a more natural narrative while DevRecap remains responsible for factual validation.
devrecap/
├── plugin.json
├── skills/
│ └── devrecap/
│ └── SKILL.md
├── .agents/
│ └── plugins/
│ └── marketplace.json
├── .claude/
│ └── skills/
│ └── devrecap/
│ └── SKILL.md
├── apps/
│ ├── cli/
│ ├── server/
│ └── web/
├── packages/
│ ├── collectors/
│ ├── activity-engine/
│ ├── report-engine/
│ └── shared/
├── scripts/
└── tests/
Developer activity is messy.
A single task may appear across:
A naive summary sees these as unrelated events.
DevRecap tries to reconstruct the workstream.
That makes the output useful for:
DevRecap is built around a few simple ideas:
Issues, ideas, integrations, report styles, collectors, and improvements are welcome.
If DevRecap becomes useful in your workflow, consider starring the repository. It helps other developers discover the project.
MIT. LICENSE
You already did the work.
DevRecap helps you remember what the evidence says you did.
78 commits
TypeScript
89.2%
JavaScript
9.2%
CSS
1.6%
Turn Codex, Claude Code and Git history into evidence-backed developer recaps, standups and work reports.
1
stars
78
commits
TypeScript
primary language
Sep 14, 2026
updated
Your coding history already knows what you did. DevRecap turns it into a recap.
Codex + Claude + Git → evidence-backed developer recaps.
Now available in the OpenAI Plugins Directory.
Install with codex plugin add devrecap@openai-curated, then run $devrecap.
You know the question.
“What did I actually work on this week?”
Then you open GitHub, scroll through commits, search old terminal sessions, try to remember what was finished, what was only investigated, and what is still in progress.
DevRecap reconstructs that work for you.
It reads only the local coding history and Git evidence you explicitly authorize, correlates the evidence, separates completed work from unfinished work, and lets the host AI turn those facts into a polished work report.
Not a timesheet. Not background monitoring. A recap you explicitly ask for when you need it.
DevRecap 0.3.0 is publicly available through the OpenAI Plugins Directory.
Install the approved public release directly from the OpenAI curated marketplace:
codex plugin add devrecap@openai-curated
Start or restart Codex, then invoke the Skill:
$devrecap
You can also install it interactively:
/plugins
Search for DevRecap, install it, start a new thread, then run:
$devrecap
If you want the latest version from this repository instead of the curated marketplace snapshot:
codex plugin marketplace add jquinteiroo/devrecap --ref main
codex plugin add devrecap@devrecap-marketplace
For most users, devrecap@openai-curated is the recommended install because it uses the reviewed public release.
Then ask naturally:
Help me remember what I worked on this week.
Or:
Me ajuda a lembrar tudo que trabalhei nas últimas duas semanas.
Quero um relatório detalhado em português.
On first use, DevRecap requires explicit permission before reading local Codex history, Claude history, or Git evidence.
DevRecap is designed to produce a recap you can actually use in a daily, weekly review, 1:1, sprint review, handoff, or performance conversation.
A typical report groups work into meaningful fronts such as:
Executive summary
Main focus
• Contract generation and document workflow
• Tournament API consistency and filtering
Confirmed deliveries
• Evidence-backed completed work
Investigations
• Work that was explored but not proven complete
In progress
• Active work without completion evidence
The final HTML report is editorial rather than commit-centric: context and narrative first, raw evidence available when you need to inspect it.
Evidence first. If DevRecap cannot support a claim from the collected evidence, it must not present it as completed.
A file edit is not automatically a finished feature.
A command is not automatically a delivered fix.
A conversation is not automatically proof that something shipped.
DevRecap combines session evidence with Git history and keeps uncertain work uncertain.
Without DevRecap:
I think I worked on the PDF generation flow...
I also fixed something in the product-selection flow.
And I investigated a bit the issues with the reports.
With DevRecap:
This week
Completed
• Improved the contract-generation flow and correlated the work with Git commits.
• Fixed frontend behavior in the product-selection journey.
Investigated
• Report-generation issues that still lacked completion evidence.
In progress
• New user creation in the admin panel.
The goal is not to make your week sound busier.
The goal is to make it accurate, useful, and easy to present.
You invoke $devrecap
↓
DevRecap resolves the requested period
↓
Authorized Codex + Claude sessions
+
Read-only Git evidence
↓
Normalize events
↓
Build meaningful Activities and Workstreams
↓
Correlate session activity with Git evidence
↓
Build an evidence-only analysis contract
↓
Codex / Claude writes the narrative
↓
DevRecap validates IDs, status and completion claims
↓
Editorial HTML report
The coding agent is the writing layer.
The DevRecap CLI is the factual layer.
That separation is intentional.
DevRecap is explicit-invocation only.
It does not automatically scan your repositories, watch your filesystem, or monitor your work in the background.
Before local collection, the user chooses which sources are allowed:
Codex history optional
Claude history optional
Git optional, read-only
Disabled sources are not invoked.
Git collection is guarded as read-only. DevRecap does not commit, push, checkout, reset, merge, rebase, or modify repositories as part of report generation.
See PRIVACY.md for the full privacy model.
The public Codex plugin ships the canonical Skill at:
skills/devrecap/SKILL.md
Claude support is also available in the repository:
.claude/skills/devrecap/SKILL.md
The AI-first flow is:
prepare
↓
.devrecap/run.json
↓
host AI analyzes contract.facts only
↓
.devrecap/analysis.json
↓
validate
↓
render
↓
reports/derecap-YYYY-MM-DD.html
The agent is instructed to:
in_progress, blocked, or unknown work to completed;DevRecap can collect authorized local Codex session history inside the requested time range.
Authorized Claude coding sessions can be included in the same recap, allowing work performed across agents to appear in one report.
DevRecap correlates relevant work with read-only Git evidence.
Commit evidence is especially useful for distinguishing:
worked on it
from:
there is evidence that this work was completed
The public directory is the easiest way to use DevRecap. Clone the repository only if you want to develop or inspect the project locally.
Clone:
git clone https://github.com/jquinteiroo/devrecap.git
cd devrecap
npm run setup
Run the CLI from the repository:
npm run recap -- "this week"
Useful commands:
npm run setup
npm test
npm run typecheck
npm run smoke
npm run dev
The direct CLI can still generate a deterministic recap without host-AI synthesis:
devrecap today
devrecap week
devrecap remember "last 14 days"
The preferred experience for polished reports is the Skill, because Codex or Claude can turn the structured Activities into a more natural narrative while DevRecap remains responsible for factual validation.
devrecap/
├── plugin.json
├── skills/
│ └── devrecap/
│ └── SKILL.md
├── .agents/
│ └── plugins/
│ └── marketplace.json
├── .claude/
│ └── skills/
│ └── devrecap/
│ └── SKILL.md
├── apps/
│ ├── cli/
│ ├── server/
│ └── web/
├── packages/
│ ├── collectors/
│ ├── activity-engine/
│ ├── report-engine/
│ └── shared/
├── scripts/
└── tests/
Developer activity is messy.
A single task may appear across:
A naive summary sees these as unrelated events.
DevRecap tries to reconstruct the workstream.
That makes the output useful for:
DevRecap is built around a few simple ideas:
Issues, ideas, integrations, report styles, collectors, and improvements are welcome.
If DevRecap becomes useful in your workflow, consider starring the repository. It helps other developers discover the project.
MIT. LICENSE
You already did the work.
DevRecap helps you remember what the evidence says you did.
78 commits
TypeScript
89.2%
JavaScript
9.2%
CSS
1.6%