session-detail
thebpandey/session-detail
A Claude skill that generates a detailed, structured session archive in markdown for long-term logging and cross-session handoff.
Instalação
npx convez-ai thebpandey/session-detail1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
session-detail
A Claude skill that records a work session verbatim instead of summarizing it. It produces a full markdown archive of everything that happened, plus a compact context-handoff block you paste into your next session to resume work without losing state.
If you have ever ended a long Claude session and then started a new one only to spend twenty minutes re-explaining what you already decided, this skill is the fix.
Table of Contents
- Why This Exists
- What It Produces
- Trigger Phrases
- Installation
- Full Example Output
- Field Reference
- How It Behaves
- Customization
- FAQ
- Author
- License
Why This Exists
Most "summary" tools compress. They throw away the exact wording of decisions, the prompts you drafted but did not run, and the small technical notes that turn out to matter three days later. That compression is the problem this skill avoids.
The distinction in one line: a summary tells you what the session was about; this archive tells you what actually happened, in enough detail to act on it later.
Three concrete situations it is built for:
- Resuming work. You stop mid-project. Next session, you paste the Context Updates block and Claude knows your locked decisions, open questions, and known gotchas immediately.
- Record-keeping. You want a dated, durable log of a build, a debugging session, or a research thread that you can drop into a repo, a wiki, or a project tracker.
- Handoff. Someone else (or future you) needs to pick up the work and needs the real detail, not a paragraph.
What It Produces
The skill always outputs exactly two sections and nothing else.
1. <session_full> (the archive, roughly 15 to 20KB)
The complete record. Includes what was accomplished, every significant decision with its rationale and alternatives, issues with root causes and resolutions, prompts that were executed, prompts drafted for the future captured verbatim, code changes, technical notes, memory state, and next-session priorities.
2. <context_updates> (the handoff, compact)
The forward-facing block. New persistent facts, changed or corrected facts, locked decisions, open questions, and carry-forward warnings. This is the part you paste into your next conversation.
Trigger Phrases
Say any of these to invoke the skill. No setup or slash command required.
session detaildetailed session summarysession archive
Installation
Option A: Claude Code (clone)
The repo name matches the skill name, so cloning produces the correct folder automatically.
Project-level (available in this project only):
cd .claude/skills/
git clone https://github.com/thebpandey/session-detail.git
User-level (available across all your projects):
cd ~/.claude/skills/
git clone https://github.com/thebpandey/session-detail.git
Either way the result is:
.claude/skills/session-detail/SKILL.md
As of the January 2026 Claude Code release, skills placed in .claude/skills activate immediately without restarting the session.
Option B: Claude.ai (zip upload)
- Download this repo as a ZIP (the green Code button, then Download ZIP), or zip the folder so that
SKILL.mdsits inside a folder namedsession-detail. - In Claude.ai, open Settings and find the custom Skills upload area, then upload the ZIP.
- Custom skill upload requires a Pro, Max, Team, or Enterprise plan with code execution enabled.
Note: the exact Settings menu label for skill upload has shifted between releases. If you do not see it, check both the Capabilities and Features areas of Settings.
Full Example Output
Below is a fabricated example showing what the skill returns. The imaginary session was a two-hour block building a tenant-screening dashboard. This is illustrative only; your real output is populated from your actual conversation.
Everything between the two horizontal rules is the skill's output. It begins immediately with the first section and ends with the single closing line. No preamble.
<session_full>
Session 2026-05-28-09-15-SESSION-01
Date: 2026-05-28 Duration: 2 hours 10 minutes
Accomplished
- Built the initial React component for the tenant-screening results table, including sortable columns for applicant name, credit band, and submission date.
- Wired the table to a mock data source so the UI can be developed before the screening API is ready.
- Added a status pill component (Pending, Approved, Flagged) with color coding driven by a single status-to-color map for maintainability.
- Fixed a column-width bug where the credit-band column collapsed when the dataset was empty.
Decisions Made
- Decision: Use a mock JSON data source for now instead of integrating the live screening API.
- Rationale: The screening API contract is not finalized, and blocking UI work on it would stall the whole sprint.
- Alternatives considered: Stubbing the API at the network layer with a service worker, which was judged heavier than needed for this stage.
- Impact: UI work can proceed independently. A swap-in task is now required once the API contract lands.
- Decision: Drive all status colors from a single map object rather than inline conditionals.
- Rationale: Status colors will be reused across the table, the detail view, and notification badges.
- Alternatives considered: Inline ternaries per component, rejected as duplicative and error prone.
- Impact: One source of truth for status styling. Adding a new status is a one-line change.
Issues Encountered
- Issue: The credit-band column collapsed to zero width when the data array was empty.
- Root cause: The column width was being inferred from content, and there was no content to infer from.
- Resolution: Set an explicit min-width on the column and added an empty-state row.
- Time spent: Approximately 25 minutes.
- Prevention: Define explicit min-widths on data-driven columns by default.
Prompts Executed
- Prompt: Generate the results table component with three sortable columns.
- Outcome: Working table with click-to-sort on all three columns.
- Key Files: src/components/ScreeningTable.jsx
- Commit: feat: add sortable screening results table
- Prompt: Add a reusable status pill driven by a color map.
- Outcome: StatusPill component plus a shared statusColors map.
- Key Files: src/components/StatusPill.jsx, src/lib/statusColors.js
- Commit: feat: add status pill with centralized color map
Prompts Generated for Next Session
Swap mock data source for live screening API
Intent: Replace the temporary mock JSON with the real screening API once its contract is finalized. Suggested filename: docs/prompts/swap-mock-for-live-api.md Target session/phase: After the API contract is signed off. Full prompt body:
Replace the mock data source in ScreeningTable.jsx with a live fetch against
the screening API. Requirements:
- Keep the existing column structure and sorting untouched.
- Add a loading state that shows skeleton rows while the request is in flight.
- Add an error state with a retry button.
- Map the API status values to the existing StatusPill statuses; do not invent
new statuses without confirming the mapping first.
- Preserve the empty-state row behavior.
Output the full updated component and a short note on any status values that
did not map cleanly.
Code Changes
- New files:
- src/components/ScreeningTable.jsx (the results table)
- src/components/StatusPill.jsx (status indicator)
- src/lib/statusColors.js (status-to-color map)
- Modified files:
- src/App.jsx (mounted the table and passed in mock data)
- Deleted files: None.
Technical Notes
- The status color map lives in src/lib/statusColors.js. Any new status must be added there first or the pill will render with a default gray.
- Sorting is client-side only. When the live API arrives, decide whether sorting moves server-side for large datasets.
- The empty-state row is keyed separately so it does not interfere with sort logic.
Memory State
- The screening API contract is not yet final. All data is currently mocked.
- Status styling is centralized and must not be duplicated inline.
- Client-side sorting is an interim choice pending dataset-size data from the live API.
Next Session
- Run the swap-mock-for-live-api prompt once the API contract is signed.
- Decide client-side versus server-side sorting based on expected row counts.
- Add the detail view that reuses the StatusPill component. </session_full>
<context_updates>
Context Updates for Next Session
New Persistent Facts
- The tenant-screening dashboard uses a mock JSON data source at this stage. The live screening API is not yet integrated.
- Status colors are centralized in src/lib/statusColors.js and reused across components.
- Table sorting is currently client-side only.
Changed or Corrected Facts
- OLD: Assumed the screening API would be ready for this sprint. NEW: The API contract is not final, so UI was built against mock data instead.
Decisions Locked
- Status styling is driven by a single color map, not inline conditionals, because the styling is reused in at least three places.
- UI development proceeds against mock data so it is not blocked by the unfinished API.
Open Questions
- Should sorting move server-side once real dataset sizes are known?
- What is the exact set of status values the live API will return, and do they map cleanly to Pending, Approved, and Flagged?
Carry-Forward Warnings
- Do not add a new status without updating statusColors.js first, or the pill silently renders gray.
- The mock-to-live swap is staged in docs/prompts and must run before any real screening data flows through the table. </context_updates>
Output ready. Session 2026-05-28-09-15-SESSION-01 archived.
Field Reference
| Section | Field | What goes in it |
|---|---|---|
| session_full | Accomplished | Concrete things built, fixed, or configured |
| session_full | Decisions Made | Each decision with rationale, alternatives, and impact |
| session_full | Issues Encountered | Problem, root cause, resolution, time spent, prevention |
| session_full | Prompts Executed | Prompts that were run, with outcomes and files touched |
| session_full | Prompts Generated for Next Session | Prompts drafted but not run, captured verbatim |
| session_full | Code Changes | New, modified, and deleted files |
| session_full | Technical Notes | Patterns, configs, and details worth preserving |
| session_full | Memory State | Durable project facts that should persist |
| session_full | Next Session | Checklist of priority tasks |
| context_updates | New Persistent Facts | Facts established this session that carry forward |
| context_updates | Changed or Corrected Facts | Superseded assumptions, in OLD then NEW form |
| context_updates | Decisions Locked | Final decisions that should not be relitigated |
| context_updates | Open Questions | Unresolved items needing answers |
| context_updates | Carry-Forward Warnings | Known risks, blockers, and gotchas |
How It Behaves
- No questions asked. On trigger, it reads the whole conversation and generates both sections immediately.
- No placeholders. If a field cannot be filled from the conversation, it is omitted or marked
N/Arather than left as a raw bracket. - Prompts are verbatim. Any prompt you drafted but did not run is captured in full, including code blocks and constraints, so it is recoverable later.
- Exact wording preserved. Code and specific copy produced during the session are not paraphrased.
- Session ID format.
YYYY-MM-DD-HH-MM-SESSION-01, derived from the timestamp of the first message.
Customization
The skill is a single SKILL.md file. To adjust it, edit that file:
- Change the closing line. Edit the Closing Line section near the bottom.
- Add or remove archive fields. Edit the FULL VERSION template block.
- Change trigger phrases. Edit the
descriptionfrontmatter and the Behavior section, then keep them consistent.
Keep the body under 500 lines for best performance, and keep the description under 1024 characters.
FAQ
Does it summarize or compress anything? No. It is a verbatim record by design. If you want a short summary, this is the wrong tool.
Will it leak placeholder brackets into my output?
No. Empty fields are omitted or marked N/A.
Does it work in both Claude Code and Claude.ai? Yes. See the two installation options above.
Can I rename it?
Yes, but the folder name must match the name field in the frontmatter, and trigger phrases must stay consistent across the frontmatter and the body.
Author
Bhaskar Pandey Founder, Almora Technology
- LinkedIn: linkedin.com/in/pandeybhaskar
- GitHub: github.com/thebpandey
- Email: bhaskar.knp@gmail.com
License
MIT. See LICENSE.
Claude Code
A Claude Code command version of this skill is available in the claude-code/ directory of this repo.
Installation
- Create the Claude Code commands directory if it does not exist:
mkdir -p ~/.claude/commands
- Copy the command file:
cp claude-code/session-detail.md ~/.claude/commands/session-detail.md
- Restart any active Claude Code session. The command will appear in autocomplete when you type
/.
Usage
/session-detail
Claude Code will pull live git state, generate the full session archive, write it to _sessions/ in your current repo, and ask whether you want the output displayed in chat. Default is no display; press Enter to skip.